A sea area digital twin method and system based on buoy and station data fusion

Abstract

The application provides a sea area digital twin method and system based on buoy and station data fusion, belongs to the related technical field of sea area digital twin, through constructing a multi-source heterogeneous marine data acquisition system, establishing a data quality evaluation mechanism and a similarity data compression mechanism, realizing intelligent fusion processing of buoy data and station data, solving the problem of low data fusion precision in traditional technology; through the self-organizing growth network structure of the marine dynamic response enhancement model, the neuron connection can be automatically adjusted according to the spatio-temporal variation degree of the training data, forming an optimal topological structure, effectively solving the defect that the traditional fixed parameter model cannot adaptively process complex marine environment; through the establishment of a dynamic parameter adjustment mechanism and a multi-algorithm adaptive switching strategy, the most suitable prediction algorithm is automatically selected according to the correlation characteristics of the marine data sequence, realizing high-precision digital reproduction of complex marine environment.

Description

Technical Field

[0001] This application relates to the field of marine digital twin technology, and in particular to a marine digital twin method and system based on the fusion of buoy and station data. Background Technology

[0002] As an important means of marine environmental monitoring and prediction, marine digital twin technology traditionally relies on a single data source or simple multi-source data overlay for marine environmental modeling. It mainly uses numerical models with fixed parameters and static data processing methods to achieve digital representation of the marine state. This type of technology has been widely used in marine environmental monitoring, marine disaster early warning, marine resource development and marine ecological protection.

[0003] However, existing technologies often fail to achieve satisfactory data fusion results when processing multi-source heterogeneous data such as buoy observation data, station monitoring data, and satellite remote sensing data. This is due to issues such as inconsistent data formats, significant differences in spatiotemporal resolution, and varying data quality. These technologies typically employ simple data averaging or interpolation methods, lacking effective data quality assessment and intelligent data processing mechanisms. Furthermore, existing marine digital twin systems struggle to adapt to the complex marine dynamics environment of the Yellow and Bohai Seas. The significant spatiotemporal heterogeneity and data quality differences among the multi-source heterogeneous observation data make it difficult to construct high-precision digital twin systems that accurately reflect the real marine environment. Summary of the Invention

[0004] This application provides a method and system for creating a marine digital twin based on the fusion of buoy and station data, aiming to solve the technical problem in the prior art that it is difficult to effectively fuse multi-source heterogeneous marine observation data to construct a high-precision digital twin system.

[0005] On the one hand, this application provides a method for creating a digital twin of a marine area based on the fusion of buoy and station data, including:

[0006] S01. Deploy an array of marine observation buoys and a network of coastal stations in the pre-defined sea area. The real-time marine parameters collected by the marine observation buoys include seawater temperature, salinity, current velocity, and wave height. The stations simultaneously monitor meteorological data and nearshore marine environmental parameters to establish a multi-source heterogeneous marine data acquisition system.

[0007] S02. Perform quality assessment on the collected buoy data and station data, calculate the data integrity coefficient and accuracy coefficient. When the integrity coefficient ∈ [0, 0.85), start the data completion algorithm to fill in the missing values. When the accuracy coefficient deviation is > 5%, start the outlier detection and correction process.

[0008] S03. The similarity coefficient between ocean parameter vectors is calculated using a similarity data compression mechanism. When the similarity coefficient ∈ [0.85, 1], the data with high similarity is compressed and stored. When the similarity coefficient ∈ [0, 0.85), the original data storage format is maintained.

[0009] S04. Construct a coupled ocean dynamics prediction model to optimize the modeling of ocean dynamic processes. The coupled ocean dynamics prediction model integrates buoy observation data and station monitoring data, corrects model parameters through physical constraints, and outputs the three-dimensional spatial distribution prediction results of ocean current field, temperature and salinity field and wave field.

[0010] S05. When the prediction error of the ocean current field exceeds the linearity requirement of 8%, the ocean current hysteresis compensation algorithm is activated to eliminate the influence of the return error. If the ocean current hysteresis error ∈ [0, 3%], the standard ocean dynamics processing mode is adopted to establish a real-time inversion mechanism for the ocean environment state.

[0011] S06. If the Euclidean distance between the ocean feature vector and the reference ocean vector in the twelve-dimensional ocean parameter space is <0.15, keep the current ocean dynamics coupled prediction model parameter configuration unchanged. When the Euclidean distance ∈ [0.15, 0.35], dynamically adjust the model convergence factor to optimize the algorithm performance.

[0012] S07. If the correlation coefficient of the ocean data sequence is >0.8, a prediction model is established using a multiple linear fitting algorithm. When the correlation coefficient of the ocean data sequence is <0.3, the piecewise fitting method is switched to handle the nonlinear relationship and a real-time mapping model of the marine digital twin system is constructed.

[0013] Optionally, the data integrity coefficient is derived from the collection and statistics of buoy data and station data. The calculation formula is: ,in The number of valid data points. This represents the theoretical total number of data points. For time coverage, The standard time coverage rate is used as the criterion for triggering the data completion algorithm.

[0014] Optionally, the accuracy coefficient is derived from the statistical deviation between the observed value and the reference standard value. The accuracy coefficient is used to assess the reliability of data quality and as the basis for initiating outlier detection. The formula for calculating the accuracy coefficient is as follows: In the formula, This is a dimensionless accuracy coefficient; The total number of data points participating in the evaluation. For the first One observation value, For the first A reference standard value, As a reference benchmark for observations, As a reference standard, The smoothing factor is calculated for accuracy.

[0015] Optionally, the similarity coefficient is derived from the cosine similarity calculation between ocean parameter vectors. The similarity coefficient is used to determine the execution conditions of the data compression strategy, wherein the calculation formula for the similarity coefficient is:

[0016] ;

[0017] In the formula, The similarity coefficient is dimensionless. Let be the dimension of the ocean parameters. and They are respectively and Time of the first One ocean parameter value, For the first Reference base values ​​for each parameter.

[0018] Optionally, the ocean dynamics coupling prediction model adopts an ocean dynamic response enhancement model. This model is based on a self-organizing growth network that automatically grows new connections between neurons according to the spatiotemporal variability of the training data, forming an optimal topology. The input layer of the ocean dynamic response enhancement model receives multidimensional ocean parameters from buoys and stations, the hidden layer uses an adaptive neural network to process the nonlinear coupling relationships of ocean dynamics, and the output layer generates three-dimensional prediction results for ocean current fields, temperature-salinity fields, and wave fields. The calculation formula for the multi-head attention mechanism parameters in the ocean dynamic response enhancement model is as follows:

[0019] ;

[0020] In the formula, These are dimensionless attention weight parameters; These are the weighting coefficients. For seawater temperature gradient, This represents the historical maximum value of the temperature gradient. For seawater salinity, This represents the historical average salinity. The standard deviation of salinity. The velocity vector magnitude, This is the highest flow rate in history.

[0021] Optionally, the ocean current hysteresis compensation algorithm calculates the hysteresis effect in the process of ocean current change by establishing a mathematical model of the hysteresis characteristics of ocean current response, and performs phase compensation on the prediction results; wherein, the ocean current hysteresis error comes from the deviation statistics between the predicted value and the actual observed value of the ocean current field, and the ocean current hysteresis error is used as the judgment condition for selecting the ocean dynamics processing mode.

[0022] The formula for calculating the ocean current hysteresis error is as follows:

[0023] ;

[0024] In the formula, This is a dimensionless ocean current hysteresis error; The number of time steps involved in the calculation. For the first Ocean current forecasts at each time step, For the first Ocean current observations at each time step, As the reference velocity for ocean currents, This is the smoothing factor for hysteresis calculations;

[0025] The calculation formula for the ocean current hysteresis compensation algorithm is as follows:

[0026] ;

[0027] In the formula, The compensated ocean current prediction value, This is the current ocean current forecast value. This is a dimensionless phase compensation coefficient. The time derivative of the ocean current forecast. The historical maximum value of the time derivative. The reference flow rate is denoted as .

[0028] Optionally, the correlation coefficient of the ocean data sequence is derived from the calculation of the Pearson correlation coefficient between time series of different ocean parameters. This correlation coefficient is used as a criterion for selecting the type of prediction algorithm. The formula for calculating the correlation coefficient of the ocean data sequence is as follows: In the formula, The correlation coefficient is dimensionless. The length of the time series. and They are respectively The two ocean parameter values ​​at time 1, and These are the time series means of the corresponding parameters. and These are the standard deviations of the corresponding parameters.

[0029] Optionally, the matrix calculation formula for the multivariate linear fitting algorithm is as follows: In the formula, The output vector has dimensions of ; The input feature matrix has a dimension of . ; The regression coefficient vector is dimensionless and has dimension 1. ; Let be the error vector, with dimension . ; These are the corresponding scaling factors.

[0030] Optionally, the segmented fitting method divides the ocean data sequence into several sub-intervals according to the trend of change, and establishes a fitting model independently for each sub-interval to handle the nonlinear and multi-stage change characteristics of the ocean system.

[0031] On the other hand, this application provides a marine digital twin system based on the fusion of buoy and station data, wherein the marine digital twin system is configured to perform the above-described marine digital twin method.

[0032] The beneficial technical effects of this application are as follows:

[0033] This application provides a method and system for marine digital twin based on buoy and station data fusion. By constructing a multi-source heterogeneous marine data acquisition system and establishing a data quality assessment mechanism and a similarity data compression mechanism, it achieves intelligent fusion processing of buoy and station data, solving the problem of low data fusion accuracy in traditional technologies. This invention utilizes the self-organizing growth network structure of the marine dynamic response enhancement model, which automatically adjusts neuron connections based on the spatiotemporal variability of training data to form an optimal topology. Combined with a marine correlation discovery function, it identifies potential correlations between events with extremely low probabilities, effectively addressing the shortcomings of traditional fixed-parameter models in adaptively handling complex marine environments. Furthermore, by establishing a dynamic parameter adjustment mechanism and a multi-algorithm adaptive switching strategy, this invention automatically selects the most suitable prediction algorithm based on the correlation characteristics of marine data sequences, achieving high-precision digital reproduction of complex marine environments. This solves the technical problem of effectively fusing multi-source heterogeneous marine observation data to construct a high-precision digital twin system in existing technologies. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart illustrating a method for creating a digital twin of a marine area based on the fusion of buoy and station data, provided for this application;

[0036] Figure 2 A schematic diagram of the self-organized growth process of a marine dynamic response enhancement model network structure using this method is provided in this application;

[0037] Figure 3 A schematic diagram illustrating the operational status of a real-time mapping model of a digital twin system for the Yellow and Bohai Seas using this method, provided in this application. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0040] It should be understood that in this application, "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product or device.

[0041] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0042] refer to Figure 1 , Figure 2 and Figure 3 As shown, this invention provides a method for creating a marine digital twin based on the fusion of buoy and station data, which solves the technical problem in existing technologies where it is difficult to effectively fuse multi-source heterogeneous marine observation data to construct a high-precision digital twin system. This invention provides a method for creating a marine digital twin based on the fusion of buoy and station data, including:

[0043] S01. Deploy an array of marine observation buoys and a network of coastal stations in the pre-defined sea area. The real-time marine parameters collected by the marine observation buoys include seawater temperature, salinity, current velocity, and wave height. The stations simultaneously monitor meteorological data and nearshore marine environmental parameters, establishing a multi-source heterogeneous marine data acquisition system.

[0044] The multi-source heterogeneous marine data acquisition system includes a collaborative acquisition architecture of buoy sensor arrays, coastal fixed stations, mobile observation platforms, and satellite remote sensing data, enabling comprehensive three-dimensional monitoring of the marine environment in the pre-defined sea area.

[0045] Specifically, a multi-source heterogeneous marine data acquisition system is constructed to achieve comprehensive, three-dimensional monitoring of the marine environment in a predetermined sea area. For example, firstly, an array of marine observation buoys is deployed in key marine areas of the predetermined sea area, with the buoy spacing controlled within a range of 10–20 km to ensure spatial continuity and representativeness of data acquisition. Each buoy is equipped with multi-parameter sensor components to collect core marine parameters such as seawater temperature, salinity, current velocity, wave height, and wind speed in real time, with a data sampling frequency set to once every 10 minutes. Simultaneously, a fixed station network is constructed along the coast, with a station density of one monitoring point every 50 km, primarily responsible for the synchronous monitoring of meteorological data and nearshore marine environmental parameters. The entire data acquisition system uses wireless communication technology to transmit data from each monitoring node to the data center in real time, and timestamp synchronization technology ensures time consistency across different data sources. The purpose of this step is to establish a multi-dimensional marine environmental monitoring network covering the Yellow and Bohai Seas, providing rich and reliable basic data support for the digital twin system.

[0046] S02. Perform quality assessment on the collected buoy data and station data, calculate the data integrity coefficient and accuracy coefficient. When the integrity coefficient ∈ [0, 0.85), start the data completion algorithm to fill in the missing values. When the accuracy coefficient deviation is > 5%, start the outlier detection and correction process.

[0047] This invention employs a data quality assessment mechanism to perform quality control and preprocessing on collected multi-source ocean data. First, a data integrity coefficient is calculated, reflecting the ratio of valid data points to the theoretical total number of data points, as well as the product of time coverage and standard time coverage. When the integrity coefficient is below 0.85, a data completion algorithm is activated, using a combination of time-series interpolation and spatial interpolation to extrapolate missing data based on ocean parameter values ​​at adjacent time points and spatial locations. Time-series interpolation uses cubic spline interpolation to handle missing data in the time dimension, while spatial interpolation uses Kriging interpolation to handle missing data in the spatial dimension. The accuracy coefficient is calculated statistically based on the deviation between observed values ​​and reference standard values. When the accuracy coefficient deviation exceeds 5%, an outlier identification method based on statistical principles is activated, using a combination of box plot rules and Z-score tests to mark and correct data deviating from the normal range. This step ensures that the ocean data entering subsequent processing flows possesses high-quality characteristics, improving the modeling accuracy of the digital twin system.

[0048] The data integrity coefficient is derived from the collection and statistics of buoy data and station data, and the calculation formula is expressed as follows: ,in The number of valid data points. This represents the theoretical total number of data points. For time coverage, The standard time coverage rate and data integrity coefficient are used as criteria to initiate the data completion algorithm. The data completion algorithm uses a combination of time series interpolation and spatial interpolation to extrapolate missing data based on oceanographic parameter values ​​at adjacent time points and spatial locations.

[0049] The accuracy coefficient is derived from the statistical deviation between observed values ​​and reference standard values. Standardization eliminates the influence of dimensions, and the accuracy coefficient is used to assess the reliability of data quality and as the basis for initiating outlier detection. Outlier detection and correction employ an outlier identification method based on statistical principles to mark and correct data that deviates from the normal range.

[0050] The data integrity coefficient is calculated using a data quality assessment mechanism to control the quality of collected multi-source ocean data.

[0051] ;

[0052] In the formula, The data integrity coefficient is a dimensionless value. The number of valid data points, expressed in units of individual data points. This represents the theoretical total number of data points, expressed in units. This refers to time coverage, measured in hours (h). Standard time coverage, in hours; This is the standard number of data points, with a default value of 10,000. For reference time length, 24 hours is used by default. The formula for calculating the accuracy coefficient is expressed as follows:

[0053] ;

[0054] In the formula, This is a dimensionless accuracy coefficient; The total number of data points participating in the evaluation, expressed in units of individual data points. For the first Each observation value, with the unit determined based on the specific parameters; For the first A reference standard value, unit and same As a reference benchmark for observations, As a reference standard, The smoothing factor, calculated for accuracy, is the historical average of the standard value. The smoothing factor is calculated for accuracy and is set to 0.01 by default. Data is obtained through real-time measurements using on-site sensors. Standard reference values ​​for the corresponding location and time can be obtained by querying standard ocean databases.

[0055] The principle behind the data integrity coefficient formula is based on data quality assessment theory. It reflects the integrity of data by multiplying the percentage of valid data by the time coverage rate. Its core calculation term is... and The product form and dimensionless processing ensure comparability between different data sources. Compared with traditional simple counting methods, this formula can comprehensively consider the spatiotemporal integrity of data and improve the accuracy and reliability of data quality assessment.

[0056] The accuracy coefficient formula uses a normalized absolute deviation calculation method, and its core term is: By using a smoothing factor to avoid cases where the denominator is zero, it has better robustness than the traditional root mean square error method and can effectively identify and quantify the accuracy level of observation data.

[0057] S03. The similarity coefficient between ocean parameter vectors is calculated using a similarity data compression mechanism. When the similarity coefficient ∈ [0.85, 1], the data with high similarity is compressed and stored. When the similarity coefficient ∈ [0, 0.85), the original data storage format is maintained.

[0058] A data compression and storage mechanism based on similarity calculation is established. By calculating the cosine similarity coefficient between ocean parameter vectors, the similarity of ocean parameters at different times or monitoring points is assessed. When the similarity coefficient is between 0.85 and 1.0, it indicates that the data has high similarity. In this case, a clustering compression algorithm is used to merge and store the ocean data with high similarity. The clustering compression algorithm is based on the k-means clustering principle, classifying similar ocean parameter vectors into the same cluster, using the cluster center to represent all data points within that cluster, achieving lossy data compression. When the similarity coefficient is below 0.85, there are significant differences in ocean parameters, and the original data storage format remains unchanged. The compression ratio is dynamically adjusted according to the similarity coefficient; the higher the similarity, the greater the compression ratio. The purpose of this mechanism is to reduce storage space usage and improve data management efficiency while ensuring data accuracy. The similarity data compression mechanism uses a clustering compression algorithm to merge and store ocean data with high similarity, reducing storage space usage while ensuring data accuracy.

[0059] The embodiments of the present invention employ a data compression and storage mechanism based on similarity calculation. The formula for calculating the similarity coefficient is expressed as follows:

[0060] ;

[0061] In the formula, The similarity coefficient is dimensionless. This represents the dimension of the ocean parameter, with a value of 12. and They are respectively and Time of the first Each ocean parameter value has a unit determined by the parameter type. For the first The reference base value for each parameter, with the same unit as the corresponding parameter. Among them, and Data is acquired through real-time monitoring of buoys and stations. The calculation method is based on the arithmetic mean of the corresponding parameter over the past year.

[0062] S04. Construct a coupled ocean dynamics prediction model to optimize ocean dynamics process modeling. The coupled ocean dynamics prediction model integrates buoy observation data and station monitoring data, corrects model parameters through physical constraints, and outputs the three-dimensional spatial distribution prediction results of ocean current field, temperature and salinity field and wave field.

[0063] This invention achieves accurate modeling and prediction of marine environmental conditions by constructing a coupled ocean dynamics prediction model. The model employs an enhanced ocean dynamic response model architecture, built upon the principle of self-organizing growth networks. The input layer of the coupled ocean dynamics prediction model receives multidimensional ocean parameters from buoys and stations, including a twelve-dimensional parameter vector encompassing seawater temperature, salinity, current velocity, and wave height. The hidden layer uses an adaptive neural network structure, dynamically adjusting the network topology to adaptively model complex nonlinear relationships in the marine environment. Neuron connections within the network dynamically grow according to an ocean correlation discovery function, which identifies potential correlations between events with extremely low probabilities. The output layer generates predictions of the three-dimensional spatial distribution of ocean currents, temperature-salinity fields, and wave fields. The coupled ocean dynamics prediction model integrates physical constraints, including mathematical expressions of the laws of conservation of mass, momentum, and energy, ensuring that the prediction results conform to the laws of ocean physics. This step establishes a mathematical model that accurately describes ocean dynamic processes, providing a core computing engine for digital twin systems.

[0064] The ocean dynamics coupled prediction model employs an ocean dynamic response enhancement model, which is based on a self-organizing growing network. This network structure automatically grows new connections between neurons based on the spatiotemporal variability of the training data, forming an optimal topology. Specifically, the ocean dynamics coupled prediction model consists of an input layer that receives multidimensional ocean parameters from buoys and stations, a hidden layer that uses an adaptive neural network to process the nonlinear coupling relationships of ocean dynamics, and an output layer that generates three-dimensional prediction results for ocean currents, temperature-salinity fields, and wave fields. The multi-head attention mechanism parameters in this ocean dynamic response enhancement model are determined based on three parameters: seawater temperature gradient, salinity distribution, and current velocity vector.

[0065] The steps for establishing the training dataset for the marine dynamic response enhancement model specifically include collecting historical marine observation data in the Yellow and Bohai Seas, including buoy monitoring records and station observation data under different seasons and meteorological conditions; performing standardized preprocessing and spatiotemporal alignment on the data; constructing a paired dataset of input feature vectors and target output vectors; and expanding the diversity of training samples through data augmentation techniques.

[0066] The specific steps for training the marine dynamic response enhancement model include initializing network weights and bias parameters, updating parameters using the backpropagation algorithm and gradient descent optimizer, evaluating model performance through cross-validation, dynamically adjusting the learning rate and regularization coefficient based on the validation error, until the marine dynamic response enhancement model converges and reaches the preset prediction accuracy requirements.

[0067] The growth of connections between neurons is determined by the ocean correlation discovery function, which helps to discover correlations between signals with extremely low probability. The growth of connections between neurons achieves adaptive modeling of complex nonlinear relationships in the ocean environment by dynamically adjusting the network topology. When a new ocean environment pattern appears, the corresponding neural connection path is automatically established, which enhances the model's ability to learn unknown ocean phenomena and improves the accuracy of the representation of multi-scale coupling processes in the ocean dynamics system.

[0068] The marine correlation discovery function is used to identify potential correlations between extremely low probability events in marine environmental parameters. Inputs include the spectral characteristics of extremely low probability signals, the time delay correlation between extremely low probability signals, the spatial distribution correlation of marine parameters, and the statistical characteristics of anomalous marine environmental events. Outputs are the adjustment coefficients for the neuron connection weights. The spectral characteristics of extremely low probability signals are derived from frequency domain analysis of marine parameter time series; the time delay correlation between extremely low probability signals is derived from the calculation of lag correlations between different marine parameters; the spatial distribution correlation of marine parameters is derived from parameter correlation analysis between different monitoring points; the statistical characteristics of anomalous marine environmental events are derived from feature extraction of historical anomalous events; and the adjustment coefficients for the neuron connection weights are used to dynamically adjust the network connection strength of the marine dynamic response enhancement model.

[0069] The physical constraints include the mathematical expressions of the laws of conservation of mass, momentum, and energy in the ocean dynamics system, used to ensure that the prediction results of the coupled ocean dynamics prediction model conform to the laws of ocean physics. The ocean current field, temperature-salinity field, and wave field are derived from the output calculations of the coupled ocean dynamics prediction model, and are used to construct a marine environmental state characterization of the digital twin system of the Yellow and Bohai Seas.

[0070] This invention constructs a coupled ocean dynamics prediction model, and the calculation formula for the multi-head attention mechanism parameters of the ocean dynamic response enhancement model is expressed as follows:

[0071] ;

[0072] In the formula, These are dimensionless attention weight parameters; These are the weighting coefficients, with default values ​​of 0.4, 0.3, and 0.3. This represents the seawater temperature gradient, expressed in °C / m. This represents the historical maximum value of the temperature gradient, in °C / m. This refers to seawater salinity, measured in psu. This is the historical average salinity, measured in psu. This represents the standard deviation of salinity, expressed in psu. The velocity vector magnitude is expressed in m / s. This represents the historical maximum flow velocity, expressed in m / s. It is obtained by calculating the difference between temperature sensor measurements from adjacent depth layers. Acquired through conductivity sensor measurement It is calculated from the three-dimensional flow velocity components measured by the flow meter.

[0073] The multi-head attention mechanism parameter formula integrates three key physical quantities—seawater temperature gradient, salinity distribution, and current velocity vector—through a weighted combination method. Dynamically adjusting attention weights, compared to fixed-weight methods, can adaptively adjust according to the actual physical state of the marine environment, thus improving the model's responsiveness to different marine phenomena.

[0074] S05. When the ocean current field prediction error exceeds the linearity requirement of 8%, the ocean current hysteresis compensation algorithm is activated to eliminate the influence of the return error. If the ocean current hysteresis error ∈ [0, 3%], the standard ocean dynamics processing mode is adopted to establish a real-time inversion mechanism for the ocean environment state.

[0075] The ocean current hysteresis compensation algorithm establishes a mathematical model of the hysteresis characteristics of ocean current response, calculates the hysteresis effect during ocean current changes, and performs phase compensation on the prediction results to eliminate prediction bias caused by seawater inertia. Ocean current hysteresis error originates from the statistical deviation between predicted and actual observed values ​​of the ocean current field, and is used as a criterion for selecting the ocean dynamics processing model. The standard ocean dynamics processing model uses conventional ocean numerical calculation methods for ocean state prediction and analysis. The real-time inversion mechanism infers the internal state parameters of the ocean environment based on current observation data, enabling real-time tracking of ocean environment changes.

[0076] The influence of backflow error in ocean current prediction is eliminated through an ocean current hysteresis compensation algorithm. First, the prediction error of the ocean current field is calculated. When the prediction error exceeds the linearity requirement of 8%, the ocean current hysteresis compensation algorithm is activated. This algorithm is based on a mathematical model of the hysteresis characteristics of ocean current response, analyzing the hysteresis effect caused by seawater inertia during ocean current changes. An ocean current response function is established to describe the input-output relationship of the ocean current system, calculating the phase hysteresis and compensating for the phase in the prediction results. When the ocean current hysteresis error is within the range of 0% to 3%, a standard ocean dynamics processing mode is adopted, using conventional ocean numerical calculation methods for ocean state prediction and analysis. Simultaneously, a real-time inversion mechanism is established to infer the internal state parameters of the ocean environment based on current observation data, achieving real-time tracking of ocean environment changes. The purpose of this step is to improve the accuracy of ocean current field prediction and reduce prediction bias caused by dynamic characteristics.

[0077] The ocean current hysteresis compensation algorithm eliminates the influence of return error in ocean current prediction. The formula for calculating ocean current hysteresis error is as follows:

[0078] ;

[0079] In the formula, This is a dimensionless ocean current hysteresis error; The number of time steps involved in the calculation, expressed in units of steps. For the first Ocean current prediction values ​​at each time step, in m / s; For the first Ocean current observations at each time step, in m / s; The reference velocity for ocean currents is the historical average velocity, expressed in m / s. This is the smoothing factor for hysteresis calculation, with a default value of 0.005. The calculation formula for ocean current hysteresis compensation is as follows:

[0080] ;

[0081] In the formula, The current is the compensated predicted value, in m / s; This is the current ocean current forecast value, in m / s; This is a dimensionless phase compensation coefficient, with a default value range of 0.1 to 0.5. The time derivative of the ocean current forecast is given in units of 1000 ppm. ; The historical maximum value of the time derivative, in units of ; As the reference flow rate, and the above They have the same meaning, and the unit is m / s. Among them, It is obtained by calculating the difference between ocean current predictions at adjacent time steps. This information was obtained through statistical analysis of historical ocean current variation rates.

[0082] The ocean current hysteresis error formula statistically quantifies the degree of hysteresis by measuring the relative deviation between predicted and observed values. Its core calculation method is... The statistical average of the ocean current hysteresis compensation formula introduces a time derivative term based on the phase correction principle. Dynamic compensation, compared to static compensation methods, can adaptively adjust according to the rate of ocean current change, effectively reducing systematic biases in ocean current prediction.

[0083] S06. If the Euclidean distance between the ocean feature vector and the reference ocean vector in the twelve-dimensional ocean parameter space is <0.15, keep the current ocean dynamics coupled prediction model parameter configuration unchanged. When the Euclidean distance ∈ [0.15, 0.35], dynamically adjust the model convergence factor to optimize the algorithm performance.

[0084] The twelve-dimensional ocean parameter space comprises twelve dimensions of marine environmental parameters, including sea surface temperature, salinity, ocean current velocity, wave height, wind speed, air pressure, tidal height, chlorophyll concentration, dissolved oxygen concentration, nutrient concentration, suspended matter concentration, and seabed topography. This space is used to calculate the Euclidean distance between the ocean feature vector and the baseline ocean vector. The ocean feature vector is derived from the real-time observed values ​​of each parameter in the twelve-dimensional ocean parameter space at the current moment, and is used for distance comparison with the baseline ocean vector. The baseline ocean vector is derived from the standard value vector of marine environmental parameters obtained through long-term observation data of the sea area, and serves as a reference benchmark for assessing the difference between the current ocean state and historical typical states. The Euclidean distance is calculated from the geometric distance between the ocean feature vector and the baseline ocean vector in the twelve-dimensional ocean parameter space, and is used to determine the parameter adjustment strategy for the ocean dynamics coupled prediction model. The model convergence factor is derived from the convergence speed control parameter during the training process of the ocean dynamics coupled prediction model, and is used to optimize the training efficiency and prediction accuracy of the model.

[0085] An adaptive adjustment mechanism for model parameters is established, and the formula for calculating the Euclidean distance in the twelve-dimensional ocean parameter space is expressed as follows:

[0086] ;

[0087] In the formula, Let be the dimensionless Euclidean distance; let be the th eigenvector of the ocean. The component, the unit of which is determined according to the parameter type; is the first component of the baseline ocean vector. Each component, unit and same; For the first The feature range of each parameter, with the same unit as the corresponding parameter, is taken as the range of historical data. Among them, The current parameter value is obtained through a real-time monitoring system. The data was obtained through historical data statistical analysis, and the calculation method was based on the median value of the corresponding parameter over the past five years. It is calculated by the difference between the maximum and minimum values ​​in historical data.

[0088] An adaptive adjustment mechanism for model parameters is established. The Euclidean distance between the ocean feature vector and the baseline ocean vector in the twelve-dimensional ocean parameter space is calculated. This distance reflects the degree of difference between the current ocean state and historical typical states. The ocean feature vector consists of real-time observed values ​​of each parameter in the twelve-dimensional ocean parameter space at the current moment. The baseline ocean vector is derived from the standard values ​​of ocean environmental parameters obtained from long-term observation data of the Yellow and Bohai Seas. When the Euclidean distance is less than 0.15, it indicates that the current ocean state is similar to the historical typical state, and the parameter configuration of the current ocean dynamics coupled prediction model remains unchanged. When the Euclidean distance is between 0.15 and 0.35, it indicates that the ocean state has changed significantly, requiring dynamic adjustment of the model convergence factor to optimize algorithm performance. The convergence factor adjustment is based on the gradient descent optimization principle, adjusting the learning rate and regularization coefficient according to the current prediction error and historical convergence trend. This step enables the model to adapt to dynamic changes in the ocean environment and maintain the stability of prediction accuracy.

[0089] S07. If the correlation coefficient of the ocean data sequence is >0.8, a prediction model is established using a multiple linear fitting algorithm. When the correlation coefficient of the ocean data sequence is <0.3, the piecewise fitting method is switched to handle the nonlinear relationship and a real-time mapping model of the marine digital twin system is constructed.

[0090] The correlation coefficient of ocean data sequences is derived from the Pearson correlation coefficient calculation between time series of different ocean parameters, and is used as a criterion for selecting the type of prediction algorithm. The multiple linear fitting algorithm uses the least squares method to establish a linear relationship model between ocean parameters, and solves for regression coefficients through matrix operations to achieve linear prediction of the ocean environment state. The piecewise fitting method divides the ocean data sequence into several sub-intervals according to the trend of change, and independently establishes a fitting model for each sub-interval to handle the nonlinear and multi-stage change characteristics of the ocean system. The real-time mapping model of the marine digital twin system establishes a dynamic correspondence between the virtual and real ocean environments, and realizes the digital reproduction of the ocean state and prediction of future trends through numerical calculation and visualization techniques. The degree of spatiotemporal variability is derived from the statistical analysis of the variation amplitude of ocean parameters in the time and spatial dimensions in the training data, and is used to guide the self-organized growth of the network structure of the ocean dynamic response enhancement model.

[0091] The appropriate prediction algorithm is selected based on the correlation characteristics of marine data. The formula for calculating the correlation coefficient of marine data sequences is as follows:

[0092] ;

[0093] In the formula, The correlation coefficient is dimensionless. This represents the length of the time series, expressed in units of time series. and They are respectively The units for the two oceanographic parameter values ​​at any given time are determined based on the parameter type. and These are the time series mean values ​​for the corresponding parameters, with the same units as the corresponding parameters. and These represent the standard deviations of the corresponding parameters, with the same units as the corresponding parameters. Among them, and Obtained through time-series data from the monitoring system. and Obtained through arithmetic mean. and The standard deviation is calculated using the standard deviation formula. The matrix expression formula for the multiple linear fitting algorithm is as follows:

[0094] ;

[0095] In the formula, The output vector has dimensions of ; The input feature matrix has a dimension of . ; The regression coefficient vector is dimensionless and has dimension 1. ; Let be the error vector, with dimension . ; These are the corresponding scaling factors, used for dimensionless processing.

[0096] The formula for calculating the criterion of extremely low probability signals is expressed as follows:

[0097] ;

[0098] In the formula, It is an indicator for judging signals with extremely low probability and no dimension. This represents the total number of signal samples, expressed in units of individual samples. For the first Each signal sample value, the unit of which is determined according to the signal type; This is the signal mean, in units of and . same; The standard deviation of the signal is expressed in units of 1 / 2π. same; This is an indicator function that takes the value 1 if the condition is met and 0 otherwise. The 99th percentile of the signal, in units of 1 / 2 and 1 / 3. same; The quantile reference value is the average of the 99th percentile of similar historical signals.

[0099] The formula for calculating the spectral entropy of a signal with extremely low probability is as follows:

[0100] ;

[0101] In the formula, The spectral entropy of a signal with extremely low probability is expressed in bits. This represents the total number of spectral components, expressed in units. For the first The power spectral density of each frequency component, in units of ; The total power spectral density is expressed in units of 1000 ppm. The calculation method is as follows .

[0102] The formula for calculating the time delay correlation between signals is as follows:

[0103] ;

[0104] In the formula, The time delay correlation is dimensionless; The number of signal pairs participating in the calculation, expressed in units of individual pairs. For the first The time delay corresponding to the maximum cross-correlation of each signal pair is expressed in seconds. For the first The maximum cross-correlation coefficient of a signal pair, dimensionless; For reference, the correlation coefficient is set to 0.5 by default.

[0105] The formula for calculating the statistical characteristic value of abnormal events is expressed as follows:

[0106] ;

[0107] In the formula, These are dimensionless statistical characteristic values ​​for anomalous events; This represents the number of abnormal events that occurred, expressed in times. This represents the total number of observation periods, expressed in units of observation time. The historical average anomaly frequency is used as the reference frequency by default. For the first The intensity index of each abnormal event, with units determined according to the event type; To reference the intensity of anomalies, the average intensity of historical anomalies is used.

[0108] The formula for calculating the ocean correlation discovery function is as follows:

[0109] ;

[0110] In the formula, The value of the correlation function is found to be dimensionless; These are weighting coefficients, with a default value of 0.25. The spectral entropy of a signal with extremely low probability is calculated using the formula above. This represents the historical maximum value of the spectral entropy, expressed in bits. The time delay correlation between signals is calculated using the formula above. The historical maximum value of the time-delay correlation is dimensionless. The spatial distribution correlation coefficient is dimensionless. This represents the maximum value of spatial correlation, and is dimensionless. These are the statistical characteristic values ​​for abnormal events; the calculation method is shown in the formula above. This represents the maximum value of the anomalous feature, and is dimensionless.

[0111] The Euclidean distance formula eliminates the influence of different parameter dimensions through standardization and adopts the root mean square form. Calculating geometric distances in multidimensional parameter spaces can more accurately reflect the actual degree of separation of parameter vectors in high-dimensional spaces compared to methods such as Manhattan distance.

[0112] The correlation coefficient formula adopts the standardized form of the Pearson correlation coefficient, through standardization transformation. and The product statistics, compared to the Spearman rank correlation coefficient, can more accurately capture the strength of the linear relationship between ocean parameters.

[0113] The matrix representation of multivariate linear fitting employs the least squares method, through matrix operations. Solving for regression coefficients is more efficient and has better numerical stability compared to methods such as stepwise regression.

[0114] The ocean correlation discovery function uses a weighted fusion of multidimensional features. It identifies the correlation between events with extremely low probability, and normalizes all items to ensure balanced contribution. Compared with single feature judgment methods, it can more comprehensively explore the implicit correlation patterns in the marine environment, providing a scientific basis for the dynamic adjustment of network structure.

[0115] The formula for judging extremely low probability signals is based on the statistical quantile theory, calculating the proportion of samples whose signal deviation exceeds the 99th quantile threshold. It is used to identify events with extremely low probability. Compared with fixed threshold methods, it has stronger adaptability and can dynamically adjust the judgment criteria according to the distribution characteristics of historical data.

[0116] The spectral entropy formula uses information entropy theory to quantify the complexity of the signal in the frequency domain by calculating the probability distribution entropy of the power spectral density of each frequency component. It reflects the spectral characteristics of a signal and can more accurately capture the frequency domain uncertainty and complexity of a signal compared to traditional power spectrum analysis methods.

[0117] The time delay correlation formula calculates the maximum cross-correlation value of multiple signal pairs and their corresponding time delays. This method quantifies the time delay correlation characteristics between signals, and can more comprehensively reflect the time correlation patterns among multiple variables in the system compared to single signal pair analysis methods.

[0118] The formula for statistical feature values ​​of abnormal events combines two dimensions: abnormal frequency and abnormal intensity. Constructing composite statistical indicators can more completely describe the statistical patterns of abnormal events compared to single-frequency statistical methods, providing a more accurate quantitative basis for anomaly pattern recognition.

[0119] Based on the correlation characteristics of ocean data, appropriate prediction algorithms are selected to construct real-time mapping models. The correlation between ocean data sequences is assessed by calculating the Pearson correlation coefficients between different ocean parameter time series. When the correlation coefficient is greater than 0.8, it indicates a strong linear relationship between ocean parameters, and a multiple linear fitting algorithm is used to establish a prediction model. This algorithm, based on the least squares principle, solves the regression coefficients through matrix operations to establish a linear relationship model between ocean parameters. When the correlation coefficient is less than 0.3, it indicates a complex nonlinear relationship between ocean parameters, and a piecewise fitting method is used. The piecewise fitting method divides the ocean data sequence into several sub-intervals according to the changing trend, and establishes a fitting model independently for each sub-interval, effectively handling the nonlinear and multi-stage changing characteristics of the ocean system. The final constructed real-time mapping model of the Yellow and Bohai Sea digital twin system establishes a dynamic correspondence between the virtual and real ocean environments, realizing the digital reproduction of ocean conditions and future trend prediction through numerical calculation and visualization techniques.

[0120] The marine dynamic response enhancement model of this invention includes a three-layer architecture. The input layer is designed as a multi-channel input structure, with each channel corresponding to a marine parameter dimension, receiving a total of twelve-dimensional marine parameter vectors, including sea surface temperature, seawater salinity, ocean current velocity, wave height, wind speed, air pressure, tidal height, chlorophyll concentration, dissolved oxygen concentration, nutrient concentration, suspended matter concentration, and seabed topography parameters. The hidden layer adopts an adaptive neural network structure. The initial network contains basic neuron nodes, and the number of nodes is determined to be 2 to 3 times the input parameter dimension. The network has self-organizing growth capabilities. When a new marine environmental pattern is detected, the system automatically generates new neuron nodes in the hidden layer and establishes corresponding connection paths. The mechanism for this growth process is based on the threshold triggering principle of the marine association discovery function. When the neuron connection weight adjustment coefficient calculated by the function exceeds the preset growth threshold of 0.75, the system determines that a new marine environmental pattern has appeared that the existing network structure cannot effectively handle. The network activation node growth algorithm first analyzes the spectral characteristic variation amplitude of extremely low probability signals to identify the location of anomalous frequency components. Then, it calculates the abrupt changes in temporal delay correlation between different ocean parameters to pinpoint parameter pairs requiring new connections. The system determines the insertion location of new neurons based on the changing patterns of spatial distribution correlation of ocean parameters, while simultaneously guiding the topology design of new connection paths by incorporating statistical characteristics of anomalous ocean environmental events. The initial weights of newly generated neurons are assigned through the output of the ocean correlation discovery function, and the connection strength is dynamically adjusted based on the correlation magnitude, forming a neural network substructure capable of responding to new ocean environmental patterns. The connection weights between neurons are dynamically adjusted through the ocean correlation discovery function, which integrates four input dimensions: spectral characteristics of extremely low probability signals, temporal delay correlation between signals, spatial distribution correlation of ocean parameters, and statistical characteristics of anomalous events. The output layer generates prediction results for three-dimensional spatial fields, corresponding to ocean current fields, temperature-salinity fields, and wave fields, with each field represented by a three-dimensional grid to indicate its spatial distribution characteristics.

[0121] The establishment of the training dataset involves four stages: historical data collection, preprocessing, feature construction, and augmentation. The historical data collection stage extracts continuous observation records from the Yellow and Bohai Sea monitoring systems over the past 5 to 10 years, covering buoy monitoring data and station observations under various marine environmental conditions, including different seasonal variations, extreme weather events, tidal cycles, and marine disasters. The preprocessing stage performs quality screening on the raw data, removing records with excessive outliers and missing values. A unified spatiotemporal grid is used to align data from different sources spatiotemporally, ensuring temporal synchronization and spatial consistency. The feature construction stage organizes the preprocessed marine parameters into paired datasets of input feature vectors and target output vectors. The input feature vectors contain the marine environmental state at the current moment and several previous time steps, while the target output vector is the predicted marine field distribution at future moments. The data augmentation stage expands the diversity of training samples and improves the model's generalization ability through techniques such as time window sliding, spatial region sampling, parameter perturbation, and synthetic data generation.

[0122] It should be noted that the key technical ideas of this invention include three core aspects: a multi-source heterogeneous data fusion mechanism, adaptive network structure optimization, and intelligent prediction under physical constraints. The multi-source heterogeneous data fusion mechanism solves the problems of scattered data sources, inconsistent formats, and varying quality in traditional marine monitoring systems by establishing a unified data quality assessment system and a similarity compression storage strategy. Compared to the simple data splicing or weighted averaging of traditional methods, this mechanism can intelligently identify data quality characteristics, prioritize the retention of high-quality data, supplement and correct low-quality data, and compress and store redundant data, thereby improving data utilization efficiency while ensuring data integrity. The adaptive network structure optimization technology, through a dynamic neuron growth mechanism driven by the marine association discovery function, enables the model to automatically adjust the network topology according to the complexity of the marine environment. Traditional neural network models use a fixed network architecture, which is difficult to adapt to the multi-scale coupling characteristics and time-varying features of marine systems. This technology, by dynamically adding neuron connections and adjusting connection weights, enables the model to have the ability to continuously learn and self-optimize, and can continuously improve its predictive performance as the marine environment changes. The physically constrained intelligent prediction technology incorporates ocean physics laws as hard constraints into the neural network learning process, ensuring that the prediction results conform to fundamental physical principles such as conservation of mass, momentum, and energy. Compared to purely data-driven machine learning methods, this technology enhances physical interpretability while maintaining prediction flexibility, avoiding prediction results that violate common physical sense.

[0123] The synergistic effect of these three key technological approaches surpasses the simple sum of individual technologies. Multi-source heterogeneous data fusion provides a high-quality training data foundation for the adaptive network structure, ensuring the network can learn realistic and effective marine environmental patterns. The dynamic optimization capability of the adaptive network structure allows physical constraints to be better integrated with data-driven learning, forming a deep fusion of physical mechanisms and intelligent algorithms. The introduction of physical constraints provides scientific guiding principles for data fusion and network optimization, avoiding blind data processing and network adjustments. These three elements mutually promote each other, forming a positive feedback loop that enables the entire digital twin system to possess self-improvement and continuous evolution capabilities, constantly enhancing prediction accuracy and applicability as marine environmental monitoring data accumulates and marine scientific understanding deepens.

[0124] This invention also addresses the technical problem of insufficient accuracy in marine environmental prediction models when handling extremely low-probability anomalies. Traditional marine prediction models are mainly based on the statistical regularities of historical data, lacking the ability to effectively identify and predict extremely low-probability marine phenomena, and failing to capture the inherent patterns of sudden and extreme events in the marine environment. This invention identifies potential correlations between extremely low-probability signals through a marine correlation discovery function, and combines this with the dynamic topology adjustment mechanism of a self-organizing growth network. When a new marine environmental pattern emerges, corresponding neural connection paths are automatically established, enabling the model to autonomously learn from unknown marine phenomena. This technical solution analyzes the spectral characteristics, temporal delay correlation, spatial distribution correlation, and statistical characteristics of extremely low-probability signals, enabling the discovery of deep correlation patterns that are difficult to identify using traditional methods, significantly improving the prediction accuracy and response capability of digital twin systems for marine anomalies.

[0125] Furthermore, this invention addresses the technical challenge of balancing computational accuracy and efficiency in ocean numerical models. Existing ocean numerical calculation methods often face the contradiction of requiring substantial computational resources for high-precision calculations while real-time requirements limit computational complexity. This is particularly true when processing large-scale ocean data, where traditional methods struggle to meet real-time processing demands while maintaining computational accuracy. This invention addresses this issue by establishing a similarity data compression mechanism and a multi-algorithm adaptive switching strategy. Based on data characteristics, it intelligently selects the optimal processing method: when ocean parameter vectors have high similarity, compressed storage is used to reduce computational load; when data sequences have strong correlations, a multivariate linear fitting algorithm is used to improve computational efficiency; and when dealing with nonlinear relationships, a piecewise fitting method is switched to ensure prediction accuracy. Simultaneously, through an ocean current hysteresis compensation algorithm and a dynamic parameter adjustment mechanism, the system can automatically optimize computational strategies under different ocean environmental conditions, ensuring both the computational accuracy of the digital twin system and meeting real-time requirements.

[0126] Specifically, the fundamental principle behind this invention's ability to solve the technical problem of constructing a high-precision digital twin system through the fusion of multi-source heterogeneous marine observation data lies in establishing a complete data quality assessment and intelligent fusion processing mechanism. Through quantitative evaluation of data integrity and accuracy coefficients, the system can automatically identify missing and abnormal data, and initiate corresponding data completion algorithms and outlier correction processing to ensure the reliability and consistency of input data. The similarity data compression mechanism, by calculating the cosine similarity between marine parameter vectors, achieves intelligent data storage optimization, ensuring both data accuracy and improved processing efficiency. The marine dynamic response enhancement model adopts a self-organizing growing network structure. Its core advantage lies in its ability to dynamically adjust the network topology according to the spatiotemporal variability of the marine environment. Through the marine correlation discovery function, it identifies correlations between events with extremely low probability that are difficult to capture using traditional methods, achieving accurate modeling of complex marine dynamic processes. The introduction of physical constraints ensures that the prediction results conform to marine physical laws, while the Euclidean distance calculation in the multi-dimensional marine parameter space and the dynamic adjustment mechanism of the model convergence factor enable the system to adaptively optimize model parameters according to actual changes in the marine environment. An adaptive switching strategy between multiple linear fitting algorithms and piecewise fitting methods automatically selects the most suitable prediction method based on different ranges of correlation coefficients in ocean data sequences. This effectively handles linear and nonlinear relationships in ocean systems and logically ensures the high-precision modeling capability of the digital twin system.

[0127] To better understand and implement this invention, an embodiment of a specific application scenario is provided below.

[0128] During the data acquisition phase, the technical team deployed 42 marine observation buoys and 18 coastal stations in the Yellow and Bohai Seas, forming a multi-source heterogeneous marine data acquisition system. The buoy sensor arrays collected real-time marine parameters such as seawater temperature, salinity, current velocity, and wave height, with a data acquisition frequency of once every 10 minutes. The coastal stations simultaneously monitored meteorological data and nearshore marine environmental parameters, including wind speed, air pressure, and tidal height. Mobile observation platforms and satellite remote sensing data served as supplementary data sources, providing the system with more comprehensive marine environmental information. During the 30-day continuous observation period, the system collected a total of 1.58 million marine environmental data points, with buoy data accounting for 67% and station data accounting for 33%.

[0129] Data quality assessment is a crucial step in ensuring system reliability. The collected buoy and station data were assessed for quality, and data integrity and accuracy coefficients were calculated. Statistical analysis revealed that the average integrity coefficient for buoy data was 0.92, and for station data, it was 0.88. When the integrity coefficient of a monitoring point fell below 0.85, the system automatically activated a data completion algorithm to fill in missing values. During the observation period, the data completion mechanism was triggered at 12 monitoring points, with missing data primarily concentrated on seawater temperature and salinity parameters. The accuracy coefficient assessment results showed that 96.3% of the data had an accuracy coefficient deviation of less than 5%, but 3.7% of the data still required outlier detection and correction.

[0130] To optimize storage efficiency, a similarity data compression mechanism was established. The system calculates the similarity coefficient between ocean parameter vectors using the cosine similarity method. Statistical results show that 78.5% of the ocean parameter vectors have a similarity coefficient above 0.85. This portion of data was merged and stored using a clustering compression algorithm, reducing storage space usage by 45%. The remaining 21.5% of the data has a similarity coefficient below 0.85, and its original data storage format was maintained to ensure data integrity.

[0131] The ocean dynamics coupled prediction model is the core component of this system. This embodiment of the invention constructs an enhanced ocean dynamic response model based on a self-organizing growing network. The input layer of this model receives multidimensional ocean parameters from 42 buoys and 18 stations. The hidden layer employs an adaptive neural network containing 256 neurons, capable of handling the nonlinear coupling relationships of ocean dynamics. The output layer generates three-dimensional prediction results for ocean currents, temperature-salinity fields, and wave fields, with a prediction time span of the next 72 hours. Model training utilizes historical oceanographic observation data from the Yellow and Bohai Seas over the past five years, including monitoring records under different seasons and meteorological conditions. The training dataset contains a total of 3.2 million sample points.

[0132] During model training, the system employs backpropagation and gradient descent optimizers for parameter updates, with an initial learning rate set to 0.001. Model performance is evaluated through cross-validation, and the root mean square error on the validation set reaches a minimum of 0.082 in the 45th training epoch. An ocean correlation discovery function is used to identify potential correlations between extremely low-probability events. This function analyzes the correlation between anomalous sea surface temperature fluctuations and storm surge events, finding a 72-hour time-delay correlation of 0.76 between the two.

[0133] Table 1 below shows the statistical characteristics of each parameter in the twelve-dimensional ocean parameter space monitored by the system:

[0134] Table 1. Statistical Characteristics of Twelve-Dimensional Ocean Parameters

[0135]

[0136] The introduction of physical constraints significantly improved the model's prediction accuracy. The system uses the mathematical expressions of the laws of conservation of mass, momentum, and energy as constraints to ensure that the prediction results conform to the laws of ocean physics. In actual operation, the average prediction error of the ocean current field was 5.2%, meeting the linearity requirements. When the prediction error of the ocean current field in a certain area exceeded 8%, the system automatically activated the ocean current hysteresis compensation algorithm, calculating the hysteresis effect by establishing a mathematical model of the hysteresis characteristics of the ocean current response. During the observation period, the hysteresis compensation mechanism was triggered 6 times, and the compensated prediction error was reduced to 2.8%.

[0137] The dynamic adjustment mechanism of model parameters ensures the system's adaptability. The Euclidean distance between the ocean feature vector and the baseline ocean vector in the 12-dimensional ocean parameter space is calculated. When the distance is less than 0.15, the current model parameter configuration remains unchanged, accounting for 73% of the total runtime. When the distance is between 0.15 and 0.35, the system dynamically adjusts the model convergence factor to optimize algorithm performance, accounting for 22% of the total runtime. When the distance is greater than 0.35, the system switches to a special processing mode, accounting for 5% of the total runtime.

[0138] Table 2 shows the types of prediction algorithms used and their performance under different correlation coefficient ranges for ocean data sequences:

[0139] Table 2 Performance Comparison of Prediction Algorithms

[0140]

[0141] The choice of prediction algorithm is based on the calculated correlation coefficients of ocean data series. Pearson correlation coefficient calculations revealed that 65% of the ocean parameter time series had correlation coefficients greater than 0.8; in these cases, the system employed a multiple linear fitting algorithm to establish the prediction model and used the least squares method to solve for the regression coefficients. For 23% of the data series with correlation coefficients between 0.3 and 0.8, a mixed fitting method was used. For 12% of the data series with correlation coefficients less than 0.3, the system switched to a piecewise fitting method to handle nonlinear relationships, dividing the data series into several sub-intervals, and establishing an independent fitting model for each sub-interval.

[0142] like Figure 2As shown, the network structure of the ocean dynamic response enhancement model can automatically grow based on the spatiotemporal variability of the training data. The spatiotemporal variability is calculated based on the statistical magnitude of changes in ocean parameters in the temporal and spatial dimensions, with an average spatiotemporal variability of 0.34. When a new ocean environmental pattern is detected, the network automatically establishes corresponding neural connection paths, growing a total of 18 new neuronal connections during the observation period. This self-organizing growth mechanism enables the model to adapt to complex changes in the ocean environment and improves its ability to learn from unknown ocean phenomena.

[0143] like Figure 3 As shown, the real-time mapping model of the Yellow and Bohai Sea digital twin system establishes a dynamic correspondence between the virtual and real marine environments. The system achieves digital reproduction of the ocean state through numerical calculation and visualization techniques, with a prediction time resolution reaching the hourly level. The real-time inversion mechanism infers the internal state parameters of the marine environment based on current observation data, enabling real-time tracking of changes in the marine environment, with an inversion accuracy of 94.6%.

[0144] The system demonstrated excellent performance in actual operation. The mean absolute error of the ocean current field prediction was 0.15. The mean absolute errors in the predicted temperature and salinity fields were 0.8℃ and 0.5℃, respectively. The mean absolute error of the wave field prediction is 0.3. The parameters of the multi-head attention mechanism are determined based on three parameters: seawater temperature gradient, salinity distribution, and current velocity vector. The dynamic adjustment of attention weights improves the model's ability to identify key ocean features.

[0145] After 30 days of continuous operation and verification, the digital twin system successfully achieved high-precision monitoring and prediction of the marine environment in the Yellow and Bohai Seas. The system's efficiency in processing marine data is 3.2 times higher than traditional methods, and its prediction accuracy is improved by 15%. In particular, the system demonstrates excellent stability and reliability when dealing with marine environmental changes under extreme weather conditions.

[0146] This invention represents a significant technological advancement over traditional marine monitoring methods. Traditional methods primarily rely on single data sources for marine environmental monitoring, exhibiting limited data fusion capabilities. In contrast, this invention achieves collaborative fusion of data from buoys, stations, mobile platforms, and satellites through a multi-source heterogeneous data acquisition system, substantially improving data integrity and reliability. Regarding data processing, traditional methods lack intelligent quality assessment mechanisms, often requiring manual intervention to handle abnormal data. This invention establishes an automated data quality assessment and anomaly detection mechanism capable of identifying and correcting problematic data in real time. In terms of prediction models, traditional marine numerical models are mostly static structures, struggling to adapt to the complex and ever-changing marine environment. This invention employs a self-organizing growing network that automatically adjusts its structure according to changes in the marine environment, significantly enhancing the model's adaptability. Regarding physical constraints, traditional methods often neglect the constraints of marine physics laws, potentially leading to predictions that violate physical laws. This invention introduces the laws of conservation of mass, momentum, and energy as hard constraints, ensuring the physical rationality of the prediction results. Furthermore, the digital twin system established by this invention achieves real-time synchronous mapping between the virtual and real marine environments, providing a novel technological means for the visualization analysis and decision support of the marine environment.

[0147] This application provides a method and system for marine digital twin based on buoy and station data fusion. By constructing a multi-source heterogeneous marine data acquisition system and establishing a data quality assessment mechanism and a similarity data compression mechanism, it achieves intelligent fusion processing of buoy and station data, solving the problem of low data fusion accuracy in traditional technologies. This invention utilizes the self-organizing growth network structure of the marine dynamic response enhancement model, which automatically adjusts neuron connections based on the spatiotemporal variability of training data to form an optimal topology. Combined with a marine correlation discovery function, it identifies potential correlations between events with extremely low probabilities, effectively addressing the shortcomings of traditional fixed-parameter models in adaptively handling complex marine environments. Furthermore, by establishing a dynamic parameter adjustment mechanism and a multi-algorithm adaptive switching strategy, this invention automatically selects the most suitable prediction algorithm based on the correlation characteristics of marine data sequences, achieving high-precision digital reproduction of complex marine environments. This solves the technical problem of effectively fusing multi-source heterogeneous marine observation data to construct a high-precision digital twin system in existing technologies.

[0148] Based on the same inventive concept, this application also provides a marine digital twin system based on buoy and station data fusion that can execute the processing flow provided in an embodiment of a marine digital twin method based on buoy and station data fusion.

[0149] This application provides a marine digital twin system based on buoy and station data fusion, which can be used to execute any of the marine digital twin methods based on buoy and station data fusion described in the above embodiments. The implementation principle and technical effects are similar and will not be repeated here. The apparatus provided in this application can be specifically used to execute the solutions provided in the corresponding method embodiments; the specific functions and achievable technical effects will not be repeated here.

[0150] This application provides a computer-readable storage medium storing computer-executable instructions. When these instructions are executed on a computer, the computer performs the aforementioned method for creating a digital twin of a sea area based on the fusion of buoy and station data.

[0151] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0152] Optionally, a readable storage medium can be coupled to the processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0153] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the above-mentioned method for creating a digital twin of a sea area based on the fusion of buoy and station data.

[0154] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for creating a digital twin of a marine area based on the fusion of buoy and station data, characterized in that, The method for digital twinning the sea area includes: S01. Deploy an array of marine observation buoys and a network of coastal stations in the pre-defined sea area. The real-time marine parameters collected by the marine observation buoys include seawater temperature, salinity, current velocity, and wave height. The stations simultaneously monitor meteorological data and nearshore marine environmental parameters to establish a multi-source heterogeneous marine data acquisition system. S02. Perform quality assessment on the collected buoy data and station data, calculate the data integrity coefficient and accuracy coefficient. When the integrity coefficient ∈ [0, 0.85), start the data completion algorithm to fill in the missing values. When the accuracy coefficient deviation is > 5%, start the outlier detection and correction process. S03. The similarity coefficient between ocean parameter vectors is calculated using a similarity data compression mechanism. When the similarity coefficient ∈ [0.85, 1], the data with high similarity is compressed and stored. When the similarity coefficient ∈ [0, 0.85), the original data storage format is maintained. S04. Construct a coupled ocean dynamics prediction model to optimize the modeling of ocean dynamic processes. The coupled ocean dynamics prediction model integrates buoy observation data and station monitoring data, corrects model parameters through physical constraints, and outputs the three-dimensional spatial distribution prediction results of ocean current field, temperature and salinity field and wave field. S05. When the prediction error of the ocean current field exceeds the linearity requirement of 8%, the ocean current hysteresis compensation algorithm is activated to eliminate the influence of the return error. If the ocean current hysteresis error ∈ [0, 3%], the standard ocean dynamics processing mode is adopted to establish a real-time inversion mechanism for the ocean environment state. S06. If the Euclidean distance between the ocean feature vector and the reference ocean vector in the twelve-dimensional ocean parameter space is <0.15, keep the current ocean dynamics coupled prediction model parameter configuration unchanged. When the Euclidean distance ∈ [0.15, 0.35], dynamically adjust the model convergence factor to optimize the algorithm performance. S07. If the correlation coefficient of the ocean data sequence is >0.8, a prediction model is established using a multiple linear fitting algorithm. When the correlation coefficient of the ocean data sequence is <0.3, the piecewise fitting method is switched to handle the nonlinear relationship and a real-time mapping model of the marine digital twin system is constructed.

2. The marine digital twin method according to claim 1, characterized in that, The data integrity coefficient is derived from the collection and statistics of buoy data and station data. The calculation formula is: ,in The number of valid data points. This represents the theoretical total number of data points. For time coverage, The standard time coverage rate is used as the criterion for triggering the data completion algorithm.

3. The marine digital twin method according to claim 1, characterized in that, The accuracy coefficient is derived from the statistical deviation between observed values ​​and reference standard values. This accuracy coefficient is used to assess the reliability of data quality and as the basis for initiating outlier detection. The formula for calculating the accuracy coefficient is as follows: In the formula, This is a dimensionless accuracy coefficient; The total number of data points participating in the evaluation. For the first One observation value, For the first A reference standard value, As a reference benchmark for observations, As a reference standard, The smoothing factor is calculated for accuracy.

4. The marine digital twin method according to claim 1, characterized in that, The similarity coefficient is derived from the cosine similarity calculation between ocean parameter vectors. This similarity coefficient is used to determine the execution conditions of the data compression strategy. The formula for calculating the similarity coefficient is as follows: ； In the formula, The similarity coefficient is dimensionless. Let be the dimension of the ocean parameters. and They are respectively and Time of the first One ocean parameter value, For the first Reference base values ​​for each parameter.

5. The marine digital twin method according to claim 1, characterized in that, The ocean dynamics coupling prediction model employs an ocean dynamic response enhancement model. This model is based on a self-organizing growth network that automatically grows new connections between neurons according to the spatiotemporal variability of the training data, forming an optimal topology. The input layer of the ocean dynamic response enhancement model receives multidimensional ocean parameters from buoys and stations, the hidden layer uses an adaptive neural network to process the nonlinear coupling relationships of ocean dynamics, and the output layer generates three-dimensional prediction results for ocean current fields, temperature and salinity fields, and wave fields. The calculation formula for the multi-head attention mechanism parameters in the ocean dynamic response enhancement model is as follows: ； In the formula, These are dimensionless attention weight parameters; These are the weighting coefficients. For seawater temperature gradient, This represents the historical maximum value of the temperature gradient. For seawater salinity, This represents the historical average salinity. The standard deviation of salinity. The velocity vector magnitude, This is the highest flow rate in history.

6. The marine digital twin method according to claim 1, characterized in that, The current hysteresis compensation algorithm establishes a mathematical model of the hysteresis characteristics of the current response, calculates the hysteresis effect in the process of current change, and performs phase compensation on the prediction results; wherein, the current hysteresis error comes from the statistical deviation between the predicted value and the actual observed value of the current field, and the current hysteresis error is used as a judgment condition for selecting the ocean dynamics processing mode. The formula for calculating the ocean current hysteresis error is as follows: ; In the formula, This is a dimensionless ocean current hysteresis error; The number of time steps involved in the calculation. For the first Ocean current forecasts at each time step, For the first Ocean current observations at each time step, As the reference velocity for ocean currents, This is the smoothing factor for hysteresis calculations; The calculation formula for the ocean current hysteresis compensation algorithm is as follows: ； In the formula, The compensated ocean current prediction value, This is the current ocean current forecast value. This is a dimensionless phase compensation coefficient. Let be the time derivative of the ocean current forecast, and be the historical maximum value of the time derivative. The reference flow rate is denoted as .

7. The marine digital twin method according to claim 1, characterized in that, The correlation coefficient of the ocean data series is derived from the Pearson correlation coefficient calculation between time series of different ocean parameters. This correlation coefficient is used as a criterion for selecting the type of prediction algorithm. The formula for calculating the correlation coefficient of the ocean data series is: ； In the formula, The correlation coefficient is dimensionless. The length of the time series. and respectively The two ocean parameter values ​​at time 1, and These are the time series means of the corresponding parameters. and These are the standard deviations of the corresponding parameters.

8. The marine digital twin method according to claim 7, characterized in that, The formula for calculating the matrix in the multivariate linear fitting algorithm is as follows: ； In the formula, The output vector has dimensions of ; The input feature matrix has a dimension of . ; The regression coefficient vector is dimensionless and has dimension 1. ; Let be the error vector, with dimension . ; These are the corresponding scaling factors.

9. The marine digital twin method according to claim 7, characterized in that, The segmented fitting method divides the ocean data sequence into several sub-intervals according to the trend of change, and establishes a fitting model independently for each sub-interval to handle the nonlinear and multi-stage change characteristics of the ocean system.

10. A marine digital twin system based on buoy and station data fusion, characterized in that, The marine digital twin system is configured to perform the marine digital twin method as described in any one of claims 1 to 9.

Images