A method and system for correcting the consistency of star-marked modes for long-term spectra

Abstract

This invention provides a method and system for constellation model consistency correction for long-term spectral data, relating to the field of ocean wave observation and forecasting technology. Through deep fusion of multi-source data, refined frequency band division, and a closed-loop feedback optimization mechanism, this invention overcomes the shortcomings of existing ocean numerical forecasting in data matching, spectral evaluation, and long-term error quantification. It overcomes the statistical uncertainty caused by the scarcity of high-quality buoys; solves the problem of balancing accuracy and sample size in single-frequency or total wave height evaluation; successfully achieves systematic quantification of the continuous decay law of long-term forecasts; and fundamentally improves the forecasting performance of numerical wave models.

Description

Technical Field

[0001] This invention relates to the field of ocean wave observation and forecasting technology, specifically to a method and system for correcting the consistency of star modes for long-term wave spectra. Background Technology

[0002] Ocean wave forecasting is a crucial foundation for marine engineering and offshore operations. With the development of remote sensing technology, satellite spectral observations have gradually become an important supplement to traditional numerical models and discrete buoy observations. How to efficiently integrate multi-source heterogeneous sea state data to improve the accuracy and spatiotemporal coverage of long-term ocean wave forecasts has become an important trend in the development of marine information technology and numerical forecasting techniques.

[0003] Existing technologies face the following challenges in data fusion and evaluation: For example, the traditional spatiotemporal matching method used by Cotton et al. is limited by the number of high-quality buoys, leading to increased statistical uncertainty in the calibration model. Assessments often employ a single frequency or total wave height, making it difficult to balance assessment accuracy with sample size. When using limited observational data as the true value for skill assessment, it is difficult to systematically quantify the continuous decay patterns of long-term forecasts. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for correcting the consistency of star-based modes for long-term spectra, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for constellation mode consistency correction for long-term spectra includes the following steps: Acquire satellite wave observation spectrum data, discrete buoy observation data, and numerical wave model forecast data for the target area; The preset wave frequency range is dynamically divided into multiple cutoff frequency bands, and the equivalent wave height of each data source in each frequency band is calculated separately. An adaptive spatiotemporal search mechanism is initiated, using numerical wave model forecast data as a spatial bridging reference, to construct the first mapping relationship for the model buoy and the second mapping relationship for the model satellite, respectively. By simultaneously analyzing the first and second mapping relationships and performing equivalent parameter substitutions, the intermediate variable parameters of the numerical wave model are removed, generating a system bias correction model that is independent of the numerical wave model and directly targets satellite observation data, and then performing the correction. The model analysis field is used as the true value to perform a gridded evaluation of the corrected long-term forecasting skills, and when the error approaches the limit asymptotic value, a compensation command is generated in reverse to dynamically optimize the forecasting parameters of the numerical wave model.

[0006] A star-marker mode consistency correction system for long-term spectra includes: The data acquisition and co-location module serves as the data transmission and spatiotemporal alignment processing center, specifically comprising a data acquisition unit, a spectrum conversion unit, and an adaptive matching unit. The data acquisition unit asynchronously downloads slope spectrum data from satellite wave observations, discrete buoy observation data, and forced field data via an external interface. The spectrum conversion unit converts the original wave number slope spectrum into a standardized frequency spectrum based on deep-water dispersion relations. The adaptive matching unit executes an expansionary co-location algorithm; when the number of valid co-location samples acquired using the set basic matching parameters is less than a preset threshold, an elastic search mechanism is triggered to automatically expand the spatial search range. The frequency band partitioning and feature extraction module, connected to the data acquisition and co-localization module, serves as the center for feature extraction and dimensionality reduction. Specifically, it includes a spectral density analysis unit, a frequency boundary generation unit, and an equivalent wave height calculation unit. The spectral density analysis unit scans the historical wave meteorological characteristics of the target sea area and calculates the energy density ratio of wind waves and swells under a specific latitude and longitude grid. The frequency boundary generation unit dynamically outputs the upper and lower frequency boundary parameters of multiple equivalent energy frequency bands based on the variance distribution. The equivalent wave height calculation unit performs energy integration operations on each dynamic frequency band and calculates the square root, compressing the high-dimensional frequency spectrum matrix into an equivalent wave height vector representing the physical characteristics of the frequency band. The deviation correction module, connected to the frequency band division and feature extraction module, is used to eliminate system errors. It includes a mapping relationship establishment unit, an anomaly protection unit, and a deviation correction calculation unit. The mapping relationship establishment unit introduces environmental factor features and uses nonlinear algorithms to train nonlinear mapping models between the model and the buoy, and between the model and the satellite, respectively. The anomaly protection unit monitors the slope, intercept, and energy attenuation rate of the input data in real time. When spectral distortion is detected or the region is determined to be an extreme weather event, a hard protection mechanism is immediately triggered, and the default safety factor is invoked. The deviation correction calculation unit executes algebraic elimination logic, combines the parameters output by the mapping relationship establishment unit, substitutes them into the correction equation, and outputs the corrected satellite wave data. The forecast evaluation module is used to evaluate the model's forecast performance. It includes a truth generation unit, a data smoothing unit, and a spatial statistics unit. The truth generation unit extracts the analysis field data of the numerical wave model as a virtual truth reference system and generates a seamless evaluation dataset. The data smoothing unit performs filtering processing based on a daily average weighted algorithm for the set long-term forecast window. The spatial statistics unit aggregates data based on geographic grids, calculates the evaluation index matrix, and measures the error limit asymptote constant. The model parameter tuning module, connected to the forecast evaluation module, is used to perform closed-loop feedback and parameter optimization. Specifically, it includes an error monitoring unit, a compensation parameter generation unit, and a parameter update unit. The error monitoring unit continuously compares the real-time forecast error with the historical asymptotic limit threshold. When an error warning is triggered, the compensation parameter generation unit extracts the deviation features of specific grids and frequency bands and compiles them into a system compensation matrix. The parameter update unit writes the compensation matrix back into the configuration of the upstream numerical wave mode to dynamically correct its forecast parameters and source term parameters. Report generation module: used to integrate evaluation results and generate comprehensive evaluation reports and visualization charts; all modules are connected through standardized interfaces, supporting parallel computing and distributed deployment.

[0007] Compared with the prior art, the beneficial effects of the present invention are: This invention overcomes the shortcomings of existing marine numerical forecasting in data matching, spectral evaluation, and long-term error quantification by deep fusion of multi-source data, refined frequency band division, and closed-loop feedback optimization mechanism. This invention introduces an adaptive spatiotemporal search mechanism; when there are insufficient matching samples, it can automatically find available overlapping points, greatly expanding the effective co-location sample library; by using a multilayer perceptron combined with environmental wind speed and wave steepness to construct a nonlinear mapping relationship, and by solving simultaneous analytical algebraic equations, it successfully removes the error introduced by the numerical wave model itself as an intermediate bridge, generating an independent correction model directly for satellite observations, reducing the uncertainty in the process of multi-source heterogeneous data fusion. This invention breaks through the limitations of static division, takes the relative balance of energy variance within the frequency band as the optimization goal, and uses the cumulative integration method to dynamically and rigorously divide the complex wave frequency range into four cutoff frequency bands covering the period from swell-dominated to wind-dominated, and calculates the equivalent wave height for each band. It retains the key physical evolution characteristics of the spectrum while avoiding the huge data overhead caused by full spectrum calculation, achieving a perfect balance between evaluation accuracy and calculation sample size. When the long-term assessment error approaches the limit attenuation asymptotic value, this invention can reverse-compile the system compensation matrix. For high-frequency errors, it automatically generates correction instructions for wind and wave input coefficients and white wave dissipation coefficients, and for low-frequency swell errors, it generates correction instructions for bottom friction coefficients. After algebraically superimposing the parameters, it simultaneously uses the Courant-Friedrich-Lyuvi condition to adaptively reduce the minimum propagation time step. This eliminates forecast bias under specific sea conditions, ensures the computational stability of the numerical solver, and ultimately achieves a fully automated closed-loop system, improving the accuracy and spatiotemporal coverage of long-term wave forecasts. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the overall method steps of the present invention; Figure 2This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0010] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0011] Example 1: Please see Figure 1 This invention provides a technical solution: a star-marker mode consistency correction method for long-term spectra, comprising the following steps: Acquire satellite wave observation spectrum data, discrete buoy observation data, and numerical wave model forecast data for the target area; The preset wave frequency range is dynamically divided into multiple cutoff frequency bands, and the equivalent wave height of each data source in each frequency band is calculated separately. An adaptive spatiotemporal search mechanism is initiated, using numerical wave model forecast data as a spatial bridging reference, to construct the first mapping relationship for the model buoy and the second mapping relationship for the model satellite, respectively. By simultaneously analyzing the first and second mapping relationships and performing equivalent parameter substitutions, the intermediate variable parameters of the numerical wave model are removed, generating a system bias correction model that is independent of the numerical wave model and directly targets satellite observation data, and then performing the correction. The model analysis field is used as the true value to perform a gridded evaluation of the corrected long-term forecasting skills, and when the error approaches the limit asymptotic value, a compensation command is generated in reverse to dynamically optimize the forecasting parameters of the numerical wave model.

[0012] The specific steps for dynamically dividing the wave frequency range into multiple cutoff frequency bands include: Historical wave energy spectral density data for a specific target sea area is obtained. The optimization objective is to achieve a relatively balanced statistical variance of wave energy within each frequency band to be divided. The preset wave frequency range is 0.056 to 0.263 Hz. This range is divided into four frequency bands: band 1 (lowest frequency), band 2 (lower-middle frequency), band 3 (medium-high frequency), and band 4 (highest frequency). For each frequency band, the band energy is calculated using the following formula: ; in, and Frequency bands The lower and upper limits of frequency, A sequence of frequency bands, frequency band frequency band energy, For frequency spectrum; when frequency band energy calculation fails or When using the default value of 0, exceptions are recorded; the equivalent wave height for each non-negative frequency band is calculated, ensuring the energy spectral density is maintained. Under physical constraints: ; in, frequency band The equivalent wave height, in meters; when hour, Set to 0; the calculation applies to both model data and observation data.

[0013] Calculate the historical wave energy spectral density across the entire frequency range Total energy within Construct the cumulative distribution function of wave energy. Solve the cumulative distribution function and the target energy ratio equation. The roots of (where k=1,2,3) are used to solve for the corresponding frequency cutoff points using numerical interpolation. ;by As boundaries, the preset wave frequency range is strictly divided into four cutoff frequency bands, each with an energy percentage of 25%, namely, the lowest frequency band 1, the low-to-medium frequency band 2, the medium-to-high frequency band 3, and the highest frequency band 4.

[0014] The specific steps to initiate the adaptive spatiotemporal search mechanism include: The initial spatial matching radius is set to 50km, and the initial time matching window is ±30 minutes. Using the grid points of the numerical wave model forecast data as a reference, the number of co-location samples overlapping with buoy or satellite observation data within the initial spatiotemporal tolerance range is counted. It is determined whether the number of co-location samples is greater than or equal to a preset statistical confidence threshold (100 in this embodiment). When the number of co-location samples is lower than the statistical confidence threshold, an expansionary search is triggered, automatically expanding the spatial matching radius in 10km increments and simultaneously widening the time matching window in 10-minute increments. This cyclical expansion process continues until the number of extracted co-location samples reaches the statistical confidence threshold or the set upper limit of the search (in this embodiment, the upper limit of the spatial matching radius is 100km, and the upper limit of the time matching window is ±60 minutes). If the sample size is still insufficient even after reaching the upper limit of the search, the reference point data is discarded to finally generate a three-way co-location dataset.

[0015] The steps to eliminate intermediate variable parameters of numerical wave models in the mapping relationship include extreme sea state protection and nonlinear fitting mechanisms: Introducing ambient wind speed and wave steepness as additional feature factors, a nonlinear machine learning algorithm is used to fit and obtain the first slope. and first intercept The first mapping relationship, and the second slope Second intercept The second mapping relationship; perform anomaly isolation judgment: monitor the regression slope parameter of the nonlinear fitting output in real time, and when the regression slope is detected to be lower than the preset minimum threshold (in this embodiment, the preset minimum threshold is preferably a<0.05), or the wind speed in the target area exceeds the extreme weather indicator threshold (i.e., 10-meter ambient wind speed). When the mapping relationship is deemed divergent and invalid, the default safety correction coefficient is forcibly locked (i.e., the mapping slope is set to 1.0 and the intercept to 0.0). The equivalent wave height mediator variable of the elimination mode is eliminated through algebraic derivation, and the correction formula is generated: ; in, For the original satellite equivalent wave height, The corrected equivalent satellite wave height; when When the denominator of the correction formula is close to 0, the default correction is used. And record any anomalies; for each frequency band By applying the appropriate correction coefficients, the corrected SWIM satellite data is obtained; the corrected SWIM satellite data is consistent with the buoy observation data in terms of statistical characteristics.

[0016] By algebraically deriving the mediating variable of the elimination mode equivalent wave height, combined with the first mapping relationship, With the second mapping relationship ,in, Equivalent wave height calculated by the model.

[0017] The calculation steps for the first and second mapping relationships are as follows: Numerical wave patterns, historical spatiotemporal matching data of discrete buoys and satellite observations from the past 3 to 5 years were extracted from the target area and surrounding sea areas to construct a training database of co-positioning samples; the dataset was divided into 70% training set, 15% validation set and 15% test set according to time series order or random sampling. Introducing the 10-meter standard ambient wind speed provided by the ERA5 forced field With the steepness of the data source itself As an additional feature factor, the jointly corrected reference wave height is used to construct the three-dimensional input feature vector. The Z-Score normalization algorithm is used to normalize the 3D input feature vectors to ensure they follow a standard normal distribution, thus eliminating the interference of dimensional differences on network gradient updates; the steepness of the data source itself... The specific calculation formula is as follows: ; in, For frequency band wave energy, It is the acceleration due to gravity. For the spectrum to cross zero period, through the zeroth moment of the spectrum With second moment The calculation is as follows: ; Multilayer perceptron was used as a nonlinear machine learning algorithm for fitting. The multilayer perceptron was set to contain two hidden layers with 64 and 32 neurons respectively. The ReLU function was used as the activation function for the hidden layers. The output layer was set to be a linear layer without an activation function, and directly output a four-dimensional dynamic coefficient vector. The loss function of the network is defined as the mean squared error function, the optimizer is Adam optimizer, and the initial learning rate is set to 0.001. The preprocessed training set is input into the multilayer perceptron for backpropagation training, and the loss value of the validation set is calculated after each round. An early stopping mechanism is introduced, and the maximum number of training iterations is set to 500 rounds. If the loss value of the validation set does not decrease for 20 consecutive rounds, the stopping condition is triggered, the training is terminated, and the weight parameters with the best performance on the validation set are saved as the final model. The real-time or predicted 3D feature vector is input into the training convergent multilayer perceptron. The first two dimensions of the four-dimensional output vector output the first slope based on the co-localization features of the pattern and the buoy. and first intercept The first mapping relationship is constructed, and the latter two dimensions output the second slope based on the co-positioning features of the model and satellites. Second intercept To construct a second mapping relationship; Perform a post-threshold review to check the physical validity of the regression coefficients. or When (i.e., indicating that the mapping relationship has degraded or failed), the default correction coefficient is used (i.e.) And mark the exception log.

[0018] The specific steps for using the model analysis field as the ground truth to perform gridded evaluation of the corrected long-term forecasting skills include: extracting the analysis field data of the numerical wave model as the virtual ground truth, and performing seamless spatiotemporal matching with the forecast field data with a forecast lead time span of 8 to 14 days; performing daily average weighting processing on the matched forecast data to filter out high-frequency oscillations; merging and spatially smoothing grouping statistics of multiple adjacent geographic grids (in this embodiment, a 5×5 grid area); calculating the normalized root mean square error and Pearson correlation coefficient within the long lead time, and extracting the limit decay asymptotic constant of the error as the forecast time progresses.

[0019] Evaluation metrics were calculated based on corrected SWIM satellite data. The corrected SWIM satellite data was co-located and matched with ESPC-E model prediction data. To maintain consistency in spatiotemporal alignment across the entire system, the time matching window was strictly maintained at ±30 minutes, consistent with the previous adaptive search baseline. Bilinear interpolation was used for spatial interpolation. When co-location failed, the time point was skipped and logged, and processing continued to the next time point. Evaluation metrics included normalized bias, normalized root mean square error, scatter exponent, and Pearson correlation coefficient. When the normalization bias and root mean square error are set to 0, anomalies are recorded; when or hour, Set to 0 and record anomalies; employ a self-analysis evaluation method, using the model analysis field as the true value, to evaluate the error of the forecast field relative to the analysis field; the self-analysis evaluation utilizes the complete spatiotemporal matching between the model analysis field and the forecast field to obtain massive co-location data (in this embodiment, massive co-location data 10...). 6 Up to 10 7 (Scale), statistical results are highly robust; analysis and evaluation indicators evolve with forecast lead time, ranging from 0 to 1080 hours (0 to 45 days); smoothing filtering is applied to the evaluation indicators: ; in, For the first The evaluation index value of the forecast lead time, The smoothed value; when or If it does not exist, use the boundary value. or .

[0020] Daily average weighting is applied to the model forecast data: ; in, The parameter value for hour h. The daily average value is centered at 12:00 UTC; the same processing is applied to the model analysis field, using tau=0 data with different operating periods; the average evaluation index is calculated within the 8 to 14-day forecast lead time range, and the time-period forecast skill is reduced to climatological level; spatial grouping statistics are performed, merging data from 5×5 grid areas, with each evaluation index value based on 625 data points (5×5 grid × 25 forecast periods).

[0021] The specific steps for dynamically optimizing the forecast parameters of the numerical wave model by generating compensation instructions in reverse include: continuously monitoring the long-term evaluation error index of each grid region; when the error index of a certain grid region is detected to be approaching the limit attenuation asymptotic value, or when abnormal distortion of the spectral energy is detected, extracting the spatiotemporal error feature vector of the region; and extracting the normalized bias scalar NBIAS contained in the spatiotemporal error feature vector through a preset interface protocol; the specific calculation formula for the normalized bias scalar NBIAS is defined as follows: ; in, For the first The model forecast equivalent wave height for each sample, This corresponds to the true value (model analysis field or satellite observation correction value). The total number of samples within the spatiotemporal window; By using a preset proportional control mapping rule, the normalized deviation scalar is transformed into a reverse compensation parameter; to achieve precise tuning, correction values ​​for different physical source term parameters are generated based on the physical characteristics of the four dynamically divided cutoff frequency bands. For frequency band 3 (mid-to-high frequency) and frequency band 4 (highest frequency): calculate the correction amount for the wind and wave energy transfer coefficient. Correction amount for white wave dissipation coefficient : ; ; in, This represents the normalized bias in the high-frequency band. and These are the set proportional gain constants; As an empirical relaxation factor, its value range is set to [0.01, 0.1]. If the value is too large, it can easily lead to numerical oscillations or even model collapse; if the value is too small, the optimization convergence will be extremely slow. This embodiment is optimized through historical sensitivity experiments. To balance optimization efficiency and model stability; For the lowest frequency band 1 and the mid-to-low frequency band 2: Since low-frequency swells are mainly affected by seabed topography, the correction amount for the bottom friction coefficient is calculated. : ; in, For the normalized bias in the low-frequency band, The bottom friction gain constant (preferred in this embodiment) ); The calculated correction values ​​for each physical source term coefficient are algebraically superimposed onto the original parameters of the source term package within the numerical wave model via the parameter interface (in this embodiment: ); in the correction amount of wind and wave energy transfer coefficient Correction amount for white wave dissipation coefficient After the change, in order to maintain the numerical stability of the numerical wave solver, the minimum propagation time step is adaptively reduced proportionally according to the critical ratio of the Courant-Friedrich-Lyuvi conditions, thereby achieving safe and multi-band closed-loop dynamic optimization of the model forecast parameters.

[0022] Integrate all assessment results, including observational assessments, self-analysis assessments, and geographic distribution analysis. Generate a comprehensive assessment report, including an executive summary, assessment methodology description, key findings, conclusions, and recommendations. Key findings include: differences in forecast skill across different frequency bands (band 4 is the best, band 1 is the worst), skill decay patterns across different forecast lead times (reaching an asymptotic value at approximately 400 hours), and skill distribution characteristics across different geographic regions (regions south of 20°S have relatively poor skills). Export assessment data in CSV or NetCDF format; generate visualizations, including scatter plots, time series plots, and geographic distribution maps, in PNG or PDF format at a resolution of 300 dpi.

[0023] The observation and evaluation includes an evaluation index system, spatiotemporal boundary parameters, and frequency band characteristic parameters. The evaluation index system includes normalized deviation, normalized root mean square error, scatter exponent, and Pearson correlation coefficient. The spatiotemporal boundary parameters include an initial spatial matching radius of 50 km, an initial temporal matching window of ±30 minutes, and a spatial interpolation method. The frequency band characteristic parameters include a frequency range of 0.056–0.263, divided into four frequency bands. The self-analysis and assessment include time axis parameters, decay characteristic parameters, and smoothing parameters; the time axis parameters include a forecast lead time of 0 to 1080 hours; the seamless matching span of the forecast field data is 8 to 14 days; the decay characteristic parameters include the asymptotic value mentioned in the report's conclusion of approximately 400 hours; and the smoothing parameters include daily average weighted processing. The geographical distribution analysis includes the region south of 20°S latitude. The spatial grid parameter is the data merging and smoothing of adjacent 5×5 grid areas. The sample size calculation parameter is based on 625 data points (i.e., 5×5 grid × 25×25 forecast periods) for each evaluation index value. The standardized parameters output by the comprehensive evaluation report include data export format, chart rendering format, and lower limit of image resolution; the data export format is CSV or NetCDF; the chart rendering format is PNG or PDF; and the lower limit of image resolution is 300 dpi.

[0024] Please see Figure 2 A star-based mode consistency correction system for long-term spectra includes: The data acquisition and co-location module serves as the data transmission and spatiotemporal alignment processing center, specifically comprising a data acquisition unit, a spectrum conversion unit, and an adaptive matching unit. The data acquisition unit asynchronously downloads slope spectrum data from satellite wave observations, discrete buoy observation data, and forced field data via an external interface. The spectrum conversion unit converts the original wave number slope spectrum into a standardized frequency spectrum based on deep-water dispersion relations. The adaptive matching unit executes an expansionary co-location algorithm; when the number of valid co-location samples acquired using the set basic matching parameters is less than a preset threshold, an elastic search mechanism is triggered to automatically expand the spatial search range. The frequency band partitioning and feature extraction module, connected to the data acquisition and co-localization module, serves as the center for feature extraction and dimensionality reduction. Specifically, it includes a spectral density analysis unit, a frequency boundary generation unit, and an equivalent wave height calculation unit. The spectral density analysis unit scans the historical wave meteorological characteristics of the target sea area and calculates the energy density ratio of wind waves and swells under a specific latitude and longitude grid. The frequency boundary generation unit dynamically outputs the upper and lower frequency boundary parameters of multiple equivalent energy frequency bands based on the variance distribution. The equivalent wave height calculation unit performs energy integration operations on each dynamic frequency band and calculates the square root, compressing the high-dimensional frequency spectrum matrix into an equivalent wave height vector representing the physical characteristics of the frequency band. The deviation correction module, connected to the frequency band division and feature extraction module, is used to eliminate system errors. It includes a mapping relationship establishment unit, an anomaly protection unit, and a deviation correction calculation unit. The mapping relationship establishment unit incorporates environmental factor features and uses nonlinear algorithms to train nonlinear mapping models between the model and the buoy, and between the model and the satellite, respectively. The anomaly protection unit monitors the slope, intercept, and energy attenuation rate of the input data in real time. It identifies spectral distortion or determines that a region is experiencing an extreme weather event (in this embodiment, the determination criterion is the observed equivalent wave height). meters, or ambient wind speed When the speed reaches 100 m / s, the hard protection mechanism is immediately triggered and the default safety factor is invoked; the deviation correction calculation unit executes the algebraic elimination logic, combines the mapping relationship to establish the parameters output by the unit, substitutes them into the correction equation, and outputs the corrected satellite wave data; The forecast evaluation module is used to evaluate the model's forecast performance. It includes a truth generation unit, a data smoothing unit, and a spatial statistics unit. The truth generation unit extracts the analysis field data of the numerical wave model as a virtual truth reference system and generates a seamless evaluation dataset. The data smoothing unit performs filtering processing based on a daily average weighted algorithm for the set long-term forecast window. The spatial statistics unit aggregates data based on geographic grids, calculates the evaluation index matrix, and measures the error limit asymptote constant. The model parameter tuning module, connected to the forecast evaluation module, is used to perform closed-loop feedback and parameter optimization. Specifically, it includes an error monitoring unit, a compensation parameter generation unit, and a parameter update unit. The error monitoring unit continuously compares the real-time forecast error with the historical asymptotic limit threshold. When an error warning is triggered, the compensation parameter generation unit extracts the deviation features of specific grids and frequency bands and compiles them into a system compensation matrix. The parameter update unit writes the compensation matrix back into the configuration of the upstream numerical wave mode to dynamically correct its forecast parameters and source term parameters. Report generation module: used to integrate evaluation results and generate comprehensive evaluation reports and visualization charts; all modules are connected through standardized interfaces, supporting parallel computing and distributed deployment.

[0025] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, Min-Max Normalization and Z-Score standardization. The technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of this invention.

[0026] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0027] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for correcting the consistency of star-marker modes for long-term spectra, characterized in that, Includes the following steps: Acquire satellite wave observation spectrum data, discrete buoy observation data, and numerical wave model forecast data for the target area; The preset wave frequency range is dynamically divided into multiple cutoff frequency bands, and the equivalent wave height of each data source in each frequency band is calculated separately. An adaptive spatiotemporal search mechanism is initiated, using numerical wave model forecast data as a spatial bridging reference, to construct the first mapping relationship for the model buoy and the second mapping relationship for the model satellite, respectively. By simultaneously analyzing the first and second mapping relationships and performing equivalent parameter substitutions, the intermediate variable parameters of the numerical wave model are removed, generating a system bias correction model that is independent of the numerical wave model and directly targets satellite observation data, and then performing the correction. The model analysis field is used as the true value to perform a gridded evaluation of the corrected long-term forecasting skills, and when the error approaches the limit asymptotic value, a compensation command is generated in reverse to dynamically optimize the forecasting parameters of the numerical wave model.

2. The method for constellation mode consistency correction for long-term spectra according to claim 1, characterized in that: The specific steps for dynamically dividing the wave frequency range into multiple cutoff frequency bands include: Historical wave energy spectral density data for a specific target sea area is obtained. With the optimization objective of achieving relative equilibrium in the variance statistics of wave energy within each frequency band to be divided, the preset wave frequency range is divided into four frequency bands: band 1 (lowest frequency), band 2 (lower-middle frequency), band 3 (medium-high frequency), and band 4 (highest frequency). For each frequency band, the band energy is calculated using the following formula: ； in, and Frequency bands The lower and upper limits of frequency, A sequence of frequency bands, frequency band frequency band energy, For frequency spectrum; when frequency band energy calculation fails or When this occurs, use the default value of 0 and log any exceptions; calculate the equivalent wave height for each non-negative frequency band: ； in, frequency band The equivalent wave height, in meters; when hour, Set to 0; the calculation applies to both model data and observation data.

3. The method for constellation mode consistency correction for long-term spectra according to claim 2, characterized in that: The specific steps to initiate the adaptive spatiotemporal search mechanism include: The initial spatial matching radius is set to 50km, and the initial time matching window is ±30 minutes. Using the grid points of the numerical wave model forecast data as a reference, the number of co-location samples that overlap with buoy or satellite observation data within the initial spatiotemporal tolerance range is counted. It is determined whether the number of co-location samples is greater than or equal to the preset statistical confidence threshold. When the number of co-location samples is lower than the statistical confidence threshold, an expansionary search is triggered, automatically expanding the spatial matching radius in steps of 10km, while simultaneously widening the time matching window in steps of 10 minutes. The cyclic expansion process continues until the number of extracted co-location samples reaches the statistical confidence threshold or reaches the set upper limit boundary of the search. If the number of samples is still insufficient after reaching the upper limit boundary of the search, the reference point data is discarded to finally generate a three-party co-location dataset.

4. The method for constellation mode consistency correction for long-term spectra according to claim 3, characterized in that: The steps to eliminate intermediate variable parameters of numerical wave models in the mapping relationship include extreme sea state protection and nonlinear fitting mechanisms: Introducing ambient wind speed and wave steepness as additional feature factors, a nonlinear machine learning algorithm is used to fit and obtain the first slope. and first intercept The first mapping relationship, and the second slope Second intercept The second mapping relationship; Anomaly isolation judgment: Real-time monitoring of the regression slope parameter of the nonlinear fitting output. When the regression slope is detected to be lower than the preset minimum threshold, or the wind speed in the target area exceeds the extreme weather indicator threshold, the current mapping relationship is determined to be divergent and invalid. At this time, the default safety correction coefficient is forcibly locked; The equivalent wave height mediating variable of the mode is eliminated through algebraic derivation, and the correction formula is generated: ； in, For the original satellite equivalent wave height, The corrected equivalent satellite wave height; when When the denominator of the correction formula is close to 0, the default correction is used. And record any anomalies; for each frequency band By applying the appropriate correction coefficients, the corrected SWIM satellite data is obtained; the corrected SWIM satellite data is consistent with the buoy observation data in terms of statistical characteristics.

5. The method for star-level mode consistency correction for long-term spectra according to claim 4, characterized in that: The calculation steps for the first and second mapping relationships are as follows: Numerical wave patterns, historical spatiotemporal matching data of discrete buoys and satellite observations from the past 3 to 5 years were extracted from the target area and surrounding sea areas to construct a training database of co-positioning samples; the dataset was divided into 70% training set, 15% validation set and 15% test set according to time series order or random sampling. Introducing the 10-meter standard ambient wind speed provided by the ERA5 forced field With the steepness of the data source itself As an additional feature factor, the jointly corrected reference wave height is used to construct the three-dimensional input feature vector. The Z-Score normalization algorithm is used to normalize the 3D input feature vectors to ensure they follow a standard normal distribution, thereby eliminating the interference of dimensional differences on the network gradient update. The steepness of the data source itself The specific calculation formula is as follows: ； in, For frequency band wave energy, It is the acceleration due to gravity. For the spectrum to cross zero period, through the zeroth moment of the spectrum With second moment The calculation is as follows: ； Multilayer perceptron was used as a nonlinear machine learning algorithm for fitting. The multilayer perceptron was set to contain two hidden layers with 64 and 32 neurons respectively. The ReLU function was used as the activation function for the hidden layers. The output layer was set to be a linear layer without an activation function, and directly output a four-dimensional dynamic coefficient vector. The loss function of the network is defined as the mean squared error function, the optimizer is Adam optimizer, and the initial learning rate is set to 0.

001. The preprocessed training set is input into the multilayer perceptron for backpropagation training, and the loss value of the validation set is calculated after each round. An early stopping mechanism is introduced, and the maximum number of training iterations is set to 500 rounds. If the loss value of the validation set does not decrease for 20 consecutive rounds, the stopping condition is triggered, the training is terminated, and the weight parameters with the best performance on the validation set are saved as the final model. The real-time or predicted 3D feature vector is input into the training convergent multilayer perceptron. The first two dimensions of the four-dimensional output vector output the first slope based on the co-localization features of the pattern and the buoy. and first intercept The first mapping relationship is constructed, and the latter two dimensions output the second slope based on the co-positioning features of the model and satellites. Second intercept To construct a second mapping relationship; Perform a post-threshold review to check the physical validity of the regression coefficients. or When this happens, use the default correction factor and mark the exception log.

6. The method for constellation mode consistency correction for long-term spectra according to claim 5, characterized in that: Evaluation metrics were calculated based on corrected SWIM satellite data. The corrected SWIM satellite data was co-located and matched with ESPC-E model prediction data. To maintain consistency in spatiotemporal alignment across the entire system, the time matching window was strictly maintained at ±30 minutes, consistent with the previous adaptive search baseline. Bilinear interpolation was used for spatial interpolation. When co-location failed, the time point was skipped and logged, and processing continued to the next time point. Evaluation metrics included normalized bias, normalized root mean square error, scatter exponent, and Pearson correlation coefficient. When the normalization bias and root mean square error are set to 0, anomalies are recorded; when or hour, Set to 0 and log the exception; A self-analysis evaluation method is adopted, using the model analysis field as the true value to evaluate the error of the forecast field relative to the analysis field. The self-analysis evaluation utilizes the complete spatiotemporal matching between the model analysis field and the forecast field to obtain massive co-location data, resulting in highly robust statistical results. The evaluation indices evolve with the forecast lead time, ranging from 0 to 1080 hours. Smoothing filters are applied to the evaluation indices. ； in, For the first The evaluation index value of the forecast lead time, The value is the smoothed value; when or If it does not exist, use the boundary value. or .

7. The method for constellation mode consistency correction for long-term spectra according to claim 6, characterized in that: Daily average weighting is applied to the model forecast data: ； in, The parameter value for hour h. The daily average value is centered at 12:00 UTC; the same processing is applied to the model analysis field, using tau=0 data with different operating periods; the average evaluation index is calculated within the 8 to 14-day forecast lead time range, reducing the time-period forecast skill to climatological levels; spatial grouping statistics are performed, merging data from 5×5 grid areas, with each evaluation index value based on 625 data points.

8. The method for constellation mode consistency correction for long-term spectra according to claim 7, characterized in that: The specific steps for dynamically optimizing the forecast parameters of the numerical wave model by generating compensation instructions in reverse include: Continuously monitor long-term evaluation error indicators for each grid region; when the error indicator of a certain grid region approaches the limit attenuation asymptotic value, or when abnormal distortion of the spectral energy is detected, extract the spatiotemporal error feature vector of the region; through a preset interface protocol, extract the normalized bias scalar NBIAS contained in the spatiotemporal error feature vector; the specific calculation formula for the normalized bias scalar NBIAS is defined as follows: ； in, For the first The model forecast equivalent wave height for each sample, For the corresponding truth value, The total number of samples within the spatiotemporal window; By using a preset proportional control mapping rule, the normalized deviation scalar is transformed into a reverse compensation parameter; to achieve precise tuning, correction values ​​for different physical source term parameters are generated based on the physical characteristics of the four dynamically divided cutoff frequency bands. For frequency band 3 (mid-to-high frequency) and frequency band 4 (highest frequency): calculate the correction amount for the wind and wave energy transfer coefficient. Correction amount for white wave dissipation coefficient : ； ； in, This represents the normalized bias in the high-frequency band. and These are the set proportional gain constants; As an empirical relaxation factor; For the lowest frequency band 1 and the mid-to-low frequency band 2: Since low-frequency swells are mainly affected by seabed topography, the correction amount for the bottom friction coefficient is calculated. : ； in, For the normalized bias in the low-frequency band, The bottom friction gain constant; The calculated corrections for each physical source term coefficient are algebraically superimposed onto the original parameters of the source term package within the numerical wave model via the parameter interface; this is done in relation to the corrections for the wind and wave energy transfer coefficients. Correction amount for white wave dissipation coefficient After the change, in order to maintain the numerical stability of the numerical wave solver, the minimum propagation time step is adaptively reduced proportionally according to the critical ratio of the Courant-Friedrich-Lyuvi conditions, thereby achieving safe and multi-band closed-loop dynamic optimization of the model forecast parameters.

9. The method for star-level mode consistency correction for long-term spectra according to claim 8, characterized in that: It also includes integrating all assessment results, including observational assessments, self-analysis assessments, and geographic distribution analysis; and generating a comprehensive assessment report, including an executive summary, a description of the assessment methodology, key findings, conclusions, and recommendations.

10. A star-marker mode consistency correction system for long-term spectra, characterized in that: For performing the method steps of any one of claims 1 to 9, including: The data acquisition and co-location module serves as the data transmission and spatiotemporal alignment processing center, specifically comprising a data acquisition unit, a spectrum conversion unit, and an adaptive matching unit. The data acquisition unit asynchronously downloads slope spectrum data from satellite wave observations, discrete buoy observation data, and forced field data via an external interface. The spectrum conversion unit converts the original wave number slope spectrum into a standardized frequency spectrum based on deep-water dispersion relations. The adaptive matching unit executes an expansionary co-location algorithm; when the number of valid co-location samples acquired using the set basic matching parameters is less than a preset threshold, an elastic search mechanism is triggered to automatically expand the spatial search range. The frequency band partitioning and feature extraction module, connected to the data acquisition and co-localization module, serves as the center for feature extraction and dimensionality reduction. Specifically, it includes a spectral density analysis unit, a frequency boundary generation unit, and an equivalent wave height calculation unit. The spectral density analysis unit scans the historical wave meteorological characteristics of the target sea area and calculates the energy density ratio of wind waves and swells under a specific latitude and longitude grid. The frequency boundary generation unit dynamically outputs the upper and lower frequency boundary parameters of multiple equivalent energy frequency bands based on the variance distribution. The equivalent wave height calculation unit performs energy integration operations on each dynamic frequency band and calculates the square root, compressing the high-dimensional frequency spectrum matrix into an equivalent wave height vector representing the physical characteristics of the frequency band. The deviation correction module, connected to the frequency band division and feature extraction module, is used to eliminate system errors. It includes a mapping relationship establishment unit, an anomaly protection unit, and a deviation correction calculation unit. The mapping relationship establishment unit introduces environmental factor features and uses nonlinear algorithms to train nonlinear mapping models between the model and the buoy, and between the model and the satellite, respectively. The anomaly protection unit monitors the slope, intercept, and energy attenuation rate of the input data in real time. When spectral distortion is detected or the region is determined to be an extreme weather event, a hard protection mechanism is immediately triggered, and the default safety factor is invoked. The deviation correction calculation unit executes algebraic elimination logic, combines the parameters output by the mapping relationship establishment unit, substitutes them into the correction equation, and outputs the corrected satellite wave data. The forecast evaluation module is used to evaluate the model's forecast performance. It includes a truth generation unit, a data smoothing unit, and a spatial statistics unit. The truth generation unit extracts the analysis field data of the numerical wave model as a virtual truth reference system and generates a seamless evaluation dataset. The data smoothing unit performs filtering processing based on a daily average weighted algorithm for the set long-term forecast window. The spatial statistics unit aggregates data based on geographic grids, calculates the evaluation index matrix, and measures the error limit asymptote constant. The model parameter tuning module, connected to the forecast evaluation module, is used to perform closed-loop feedback and parameter optimization. Specifically, it includes an error monitoring unit, a compensation parameter generation unit, and a parameter update unit. The error monitoring unit continuously compares the real-time forecast error with the historical asymptotic limit threshold. When an error warning is triggered, the compensation parameter generation unit extracts the deviation features of specific grids and frequency bands and compiles them into a system compensation matrix. The parameter update unit writes the compensation matrix back into the configuration of the upstream numerical wave mode to dynamically correct its forecast parameters and source term parameters. Report generation module: used to integrate evaluation results and generate comprehensive evaluation reports and visualization charts; all modules are connected through standardized interfaces, supporting parallel computing and distributed deployment.

Images