Deep sea wind field inversion correction method, system and device based on spatio-temporal coupling of satellite and sea-based multi-source data and medium

Abstract

The application discloses a deep-sea wind field inversion correction method and system based on satellite and sea-based multi-source data space-time coupling, a device and a medium, and belongs to the technical field of sea wind field inversion correction. The method comprises the following steps: acquiring a satellite remote sensing image and a sea-based measured wind speed, obtaining an initial wind speed field through a geophysical model function inversion, constructing a space-time registration sample set with a sea-based coordinate as an anchor point, extracting a dynamic characteristic parameter of wind speed deviation, a wind direction sector acceleration ratio and a single-point continuous feature, constructing a multi-dimensional wind speed correction model based on the same, determining a nonlinear mapping relationship and generating a correction coefficient matrix, extrapolating the single-point feature to a grid in combination with the acceleration ratio, performing time interpolation and space correction on the initial wind field, and generating high-precision fusion correction wind field data. Through the multi-dimensional correction and space-time extrapolation technology, the application significantly eliminates the inversion deviation, realizes high space-time resolution reconstruction of the deep-sea wind field, and greatly improves the accuracy and continuity of the wind field data.

Description

Technical Field

[0001] This application belongs to the field of marine wind field inversion and correction technology, and specifically relates to a method, system, equipment and medium for deep-sea wind field inversion and correction based on the spatiotemporal coupling of satellite and sea-based multi-source data. Background Technology

[0002] Offshore wind power development has created an urgent need for high-precision, high-spatiotemporal-resolution nearshore wind field data. Currently, the acquisition of sea surface wind field data mainly relies on two methods: satellite remote sensing observation and sea-based field observation. Satellite remote sensing, especially spaceborne synthetic aperture radar (SAR), can provide large-scale, high spatial resolution sea surface wind field data, making up for the limited observation range of traditional shore-based observation stations, ships, and buoys. However, satellite remote sensing data has obvious drawbacks: first, the temporal resolution is low and the satellite transit period is long, making it difficult to obtain continuous, real-time wind field data; second, due to the limitations of geophysical model function accuracy and atmospheric environmental influences, the wind field data retrieved by satellite has systematic errors, limiting the retrieval accuracy. While sea-based observation methods (such as wind towers and lidar) can acquire continuous and high-precision measured wind speed data at a single point, their spatial representativeness is extremely limited. They can only reflect the wind field characteristics of a local area and cannot meet the needs of large-scale wind resource assessment. Existing wind field data processing methods often use the two data sources mentioned above in isolation, resulting in wind field data that is either spatially continuous but lacks accuracy and is temporally discrete, or highly accurate and temporally continuous but has an extremely limited spatial range.

[0003] Therefore, there is an urgent need for a method that can deeply integrate satellite wide-area discrete data with sea-based local continuous data to eliminate systematic errors in satellite inversion and achieve high-precision continuous reconstruction of wind field data in both time and space dimensions. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a method, system, device, and medium for inverting and correcting deep-sea wind fields based on the spatiotemporal coupling of satellite and sea-based multi-source data. This method can deeply fuse wide-area discrete satellite data with local continuous sea-based data to eliminate systematic errors in satellite inversion and achieve high-precision continuous reconstruction of wind field data in both time and space dimensions.

[0005] Firstly, this application provides a method for inverting and correcting deep-sea wind fields based on the spatiotemporal coupling of satellite and sea-based multi-source data, the method comprising, Acquire satellite remote sensing images of the target sea area and sea-based measured wind speeds; The satellite remote sensing image is preprocessed to obtain the backscattering coefficient and incident angle. Combined with auxiliary wind direction data, a preset geophysical model function is input to obtain the initial satellite inversion wind speed field. Using the sea-based observation spatial coordinates as anchor points, the corresponding pixel wind speed sequence is extracted from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set. The dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector are extracted from the spatiotemporal registration sample set, the wind speed acceleration ratio under each wind direction sector is calculated, and the continuous wind speed features of a single point on the sea are extracted. Based on the dynamic characteristic parameters and using the sea-based measured wind speed as a reference, a multi-dimensional wind speed correction model is constructed to determine the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and to generate a wind speed correction coefficient matrix. The target sea area is divided into grids. Using the wind speed correction coefficient matrix and the wind speed acceleration ratio, the continuous wind speed characteristics of the sea-based single point are extrapolated to the grid space. The initial satellite-retrieved wind speed field is interpolated in time and spatially corrected to generate fused and corrected wind field data.

[0006] Furthermore, The preprocessing of the satellite remote sensing image to obtain the backscattering coefficient and incident angle, combined with auxiliary wind direction data and input into a preset geophysical model function, yields the initial satellite inversion wind speed field, specifically including: The backscattering coefficient and the incident angle are obtained by performing radiometric calibration and polarization correction on the satellite remote sensing image; By combining auxiliary wind direction data with the geophysical model function, the initial satellite-retrieved wind speed field is obtained.

[0007] Furthermore, The process of extracting corresponding pixel wind speed sequences from the initial satellite-retrieved wind speed field using sea-based observation spatial coordinates as anchor points to construct a spatiotemporal registration sample set specifically includes: Using the sea-based observation spatial coordinates as anchor points, wind speed sequences of pixels at the same location in the initial satellite inversion wind speed field are extracted, and these sequences are paired with the sea-based measured wind speeds at the corresponding times to construct the spatiotemporal registration sample set.

[0008] Furthermore, The process of extracting dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculating the wind speed acceleration ratio under each wind direction sector, and extracting continuous wind speed features at a single point on the sea basis specifically includes: The satellite-retrieved wind speed is extracted from the spatiotemporal registration sample set and the difference between the retrieved wind speed and the sea-based measured wind speed is calculated as the wind speed deviation. The mean and variance of the wind speed deviation under different wind speed and wind direction sectors are statistically analyzed as the dynamic feature parameters. According to the wind direction sector corresponding to the spatiotemporal registration sample set, the sea-based measured wind speed is grouped and statistically analyzed. The sea-based measured average wind speed under each wind direction sector and the sea-based measured global average wind speed for all wind directions are calculated. The ratio of the sea-based measured average wind speed to the sea-based measured global average wind speed is used as the wind speed acceleration ratio. Based on the time series corresponding to the spatiotemporal registration sample set, the time series curve and rate of change of the measured wind speed at the sea base are extracted as the continuous wind speed characteristics of the sea base at a single point.

[0009] Furthermore, Based on the dynamic characteristic parameters and using the sea-based measured wind speed as a reference, a multi-dimensional wind speed correction model is constructed to determine the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and to generate a wind speed correction coefficient matrix, specifically including: Using the measured wind speed at sea as a reference, the mean and variance of wind speed deviation in the dynamic characteristic parameters are used as input nodes, and the ratio of satellite-retrieved wind speed to measured wind speed at sea is used as the target output node to construct a multilayer perceptron. The dynamic feature parameters are input into the multilayer perceptron, and the network weights are iteratively adjusted through the backpropagation algorithm until the loss function converges, thereby completing the construction of the multidimensional wind speed correction model to determine the nonlinear mapping relationship. The multi-dimensional wind speed correction model is trained by inputting dynamic feature parameters under different combinations of wind speed magnitude and wind direction sector. The wind speed ratio output by the model is used as the correction coefficient, and the wind speed correction coefficient matrix is ​​generated by arranging the wind speed magnitude and wind direction sector according to the combination relationship.

[0010] Furthermore, The process of dividing the target sea area into grids, using the wind speed correction coefficient matrix and the wind speed acceleration ratio, extrapolating the continuous wind speed characteristics of a single point on the sea base to the grid space, and performing time interpolation and spatial correction on the initial satellite-retrieved wind speed field to generate fused and corrected wind field data specifically includes: The wind speed correction coefficient matrix is ​​multiplied pixel by pixel by the initial satellite-inverted wind speed field to correct spatial distribution deviation; The wind speed acceleration ratio is used as a spatial scaling factor to weight and allocate the continuous wind speed characteristics of the sea-based single point and map them to a two-dimensional grid space; Spline interpolation is used to continuously fill in the time dimension of the satellite observation periods to generate the fused and corrected wind field data.

[0011] Secondly, based on the same inventive concept, this application provides a deep-sea wind field inversion and correction system based on the spatiotemporal coupling of satellite and sea-based multi-source data. The system includes: The data acquisition module acquires satellite remote sensing images of the target sea area and sea-based measured wind speeds; The initial inversion module preprocesses the satellite remote sensing image to obtain the backscattering coefficient and incident angle, and combines the auxiliary wind direction data with a preset geophysical model function to obtain the initial satellite inversion wind speed field. The feature extraction module uses the sea-based observation spatial coordinates as anchor points to extract the corresponding pixel wind speed sequence from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set. It also extracts dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculates the wind speed acceleration ratio under each wind direction sector, and extracts continuous wind speed features at a single point on the sea. The model building module, based on the dynamic feature parameters and using the sea-based measured wind speed as a reference, constructs a multi-dimensional wind speed correction model, determines the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and generates a wind speed correction coefficient matrix. The wind field fusion module divides the target sea area into grids, uses the wind speed correction coefficient matrix and the wind speed acceleration ratio to extrapolate the continuous wind speed characteristics of the sea-based single point to the grid space, performs time interpolation and spatial correction on the initial satellite-inverted wind speed field, and generates fused and corrected wind field data.

[0012] Thirdly, this application also provides an electronic device, including at least one processor and at least one memory electrically connected; The memory is electrically connected to the processor, wherein the memory stores instructions that can be executed by at least one of the processors, the instructions being executed by at least one of the processors to enable at least one of the processors to perform any of the deep-sea wind field inversion corrections based on the spatiotemporal coupling of satellite and sea-based multi-source data as described above.

[0013] Fourthly, this application also provides a computer storage medium, wherein a computer program is stored within the computer-readable storage medium; When the computer program is executed by the processor, it implements any of the above-described methods for inverting and correcting deep-sea wind fields based on the spatiotemporal coupling of satellite and sea-based multi-source data.

[0014] Fifthly, this application also provides a computer program product, which is stored in at least one storage medium; The computer program product includes several instructions to cause at least one electronic device to perform any of the deep-sea wind field inversion corrections based on the spatiotemporal coupling of satellite and sea-based multi-source data as described above.

[0015] Compared with the prior art, this application has the following advantages: 1. By constructing a spatiotemporal registration sample set of satellite discrete wind speed and sea-based continuous wind speed, and extracting dynamic characteristic parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector, a multi-dimensional wind speed correction model is constructed using sea-based measured wind speed as a reference benchmark. This model can accurately capture and compensate for systematic errors caused by factors such as the accuracy of geophysical model functions and atmospheric environment during satellite inversion, significantly improving the data accuracy of satellite remote sensing wind fields. 2. By extracting continuous wind speed features at a single point on the sea surface and extrapolating them to a two-dimensional grid space using the wind speed acceleration ratio, and by using spline interpolation to fill in the continuous time dimension during periods when the satellite is not observed, the defects of low temporal resolution of satellite remote sensing data and limited spatial range of sea-based observation data are effectively overcome, and continuous monitoring of deep-sea wind fields in time and wide-area coverage in space are achieved. 3. When extracting dynamic feature parameters, the mean and variance of wind speed deviation under different wind speed and direction sectors were statistically analyzed, and a multilayer perceptron was constructed to determine the nonlinear mapping relationship. This multi-dimensional feature quantization method enables the wind speed correction model to adapt to the error distribution under different wind speed ranges and wind direction conditions, avoiding overcompensation or undercompensation caused by a single correction coefficient, and enhancing the robustness and generalization ability of the correction model. 4. During the spatial extrapolation process, the wind speed correction coefficient matrix is ​​multiplied pixel by pixel with the initial satellite-inverted wind speed field to correct spatial deviations. At the same time, the wind speed acceleration ratio is used as a spatial scaling factor to weight and allocate the continuous wind speed characteristics on the sea basis. This takes into account both the spatial macroscopic distribution characteristics of the satellite-inverted wind field and the local fine changes of the measured wind field on the sea basis, and realizes high-fidelity reconstruction of the spatial distribution of wind fields over a large area. 5. By using a spatiotemporal coupling mechanism to deeply integrate the wide-area spatial information of satellites with the continuous temporal information of sea-based systems, fused and corrected wind field data with both high spatial resolution and high temporal continuity is generated. This provides high-precision, high spatiotemporal resolution data support for the refined assessment of deep-sea wind energy resources, micro-site selection of wind farms, and prediction of power generation. It effectively solves the problem that existing single observation methods cannot meet the actual needs of deep-sea wind power development.

[0016] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the deep-sea wind field inversion and correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data according to an embodiment of this application is shown. Figure 2 A schematic diagram of the structure of the deep-sea wind field inversion and correction system based on the spatiotemporal coupling of satellite and sea-based multi-source data according to an embodiment of this application is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Figure 1 This paper illustrates a method for inverting and correcting deep-sea wind fields based on the spatiotemporal coupling of satellite and sea-based multi-source data, according to an embodiment of this application. For example... Figure 1 As shown in the figure, the deep-sea wind field inversion and correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data in this application includes the following steps: S1, acquire satellite remote sensing images of the target sea area and sea-based measured wind speed; S2, preprocess the satellite remote sensing image to obtain the backscattering coefficient and incident angle, and combine the auxiliary wind direction data with the preset geophysical model function to obtain the initial satellite inversion wind speed field; S3, using the sea-based observation spatial coordinates as anchor points, extract the corresponding pixel wind speed sequence from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set, and extract the dynamic feature parameters of wind speed deviation as wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculate the wind speed acceleration ratio under each wind direction sector, and extract the continuous wind speed features of a single point on the sea. S4. Based on the dynamic characteristic parameters and using the sea-based measured wind speed as a reference, a multi-dimensional wind speed correction model is constructed to determine the nonlinear mapping relationship between the satellite-retrieved wind speed and the sea-based measured wind speed, and to generate a wind speed correction coefficient matrix. S5, the target sea area is divided into grids, and the continuous wind speed characteristics of the sea-based single point are extrapolated to the grid space using the wind speed correction coefficient matrix and the wind speed acceleration ratio. The initial satellite-inverted wind speed field is interpolated in time and spatially corrected to generate fused corrected wind field data.

[0021] In the specific implementation process, for the selected deep-sea target area, raw secondary product image data of spaceborne synthetic aperture radar (SAR) covering the area is acquired. The secondary product refers to the image data generated after preprocessing such as radiometric calibration and geometric positioning. It is the basic data input for subsequent wind field inversion, rather than the raw zero-order (level 0) signal directly collected by the sensor. The satellite remote sensing image has the characteristics of large-scale spatial coverage but temporal discreteness. Simultaneously, continuous high-frequency wind speed data collected by sea-based observation equipment (such as wind towers, floating lidar, etc.) deployed in the target sea area is acquired. The sea-based measured wind speed has the characteristics of single-point spatial local distribution but temporal continuity. The acquired raw sea-based wind speed data is subjected to quality control, outliers and missing values ​​are removed, and the wind speed is uniformly converted to a standard height above the sea surface (such as 10 meters) to form a temporally continuous and standardized sea-based measured wind speed sequence.

[0022] In the embodiments of this application, step S2 specifically includes: S21, Perform radiometric calibration and polarization correction on the satellite remote sensing image to obtain the backscattering coefficient and the incident angle; S22, combined with auxiliary wind direction data input into the geophysical model function, the initial satellite inversion wind speed field is obtained by inversion.

[0023] In the specific implementation process, the metadata information of the satellite remote sensing image is read, and the original digital quantization value (DN value) of the image is converted into radar backscattering coefficient using the radiometric calibration formula. In this process, the local incident angle of the radar beam is calculated pixel by pixel based on the satellite orbit parameters, sensor parameters and slant range information. For satellite data imaged using horizontal polarization (HH polarization), a polarization correction model is introduced to correct and convert the backscattering coefficient under horizontal polarization to the equivalent value under vertical polarization (VV polarization) to eliminate the influence of polarization on microwave scattering characteristics, thereby obtaining an accurate dataset of backscattering coefficients and spatially distributed incident angles. Numerical weather forecast (NWP) data or atmospheric reanalysis data covering the target sea area are acquired, and initial wind direction information matching the satellite imaging time is extracted as auxiliary wind direction data. The polarization-corrected vertical polarization backscattering coefficient, local incident angle, and the auxiliary wind direction data are substituted into a preset geophysical model function (such as the CMOD series model function) for solution. The wind speed at 10 meters above sea level corresponding to each pixel is inverted through iterative calculation or table lookup method, thereby generating an initial satellite inversion wind speed field covering the target sea area.

[0024] In the embodiments of this application, step S3 specifically includes: S311, using the sea-based observation spatial coordinates as anchor points, extract the wind speed sequence of pixels at the same location in the initial satellite inversion wind speed field, and pair the sequence with the sea-based measured wind speed at the corresponding time to construct the spatiotemporal registration sample set.

[0025] In the specific implementation process, the precise geographic latitude and longitude coordinates of the sea-based observation equipment (such as wind towers or floating lidar) are obtained and used as spatial anchor points. Based on the coordinates of the anchor points, the location is performed in the geospatial grid of the initial satellite inversion wind speed field. Considering the spatial resolution of satellite pixels and geographic registration errors, the wind speed value within a preset neighborhood window (such as a 3×3 or 5×5 pixel window) centered on the anchor point is extracted. The arithmetic mean of the wind speeds of all effective pixels within the window is calculated and used as the satellite inversion wind speed corresponding to the location of the spatial anchor point. Based on the imaging times of the target sea area over which the satellite passes, the satellite inversion wind speeds at the corresponding times are extracted to form a wind speed sequence. At the same time, in the time series of sea-based measured wind speeds, based on the satellite imaging time, sea-based measured wind speed data within a preset time window (e.g., ±10 minutes) are selected, and the average sea-based wind speed within this time period is calculated to eliminate the pulsating noise of sea-based high-frequency observations and make it match the instantaneous satellite observations on a time scale. The satellite-retrieved wind speed extracted at the same imaging time is aligned and paired with the sea-based measured wind speed after time matching to form a satellite-sea-based wind speed data pair. The spatiotemporal registration sample set is constructed by combining the data pairs of all satellite transit times.

[0026] In the embodiments of this application, step S3 further includes: S321, extract the satellite-retrieved wind speed from the spatiotemporal registration sample set and calculate the difference between the two with the sea-based measured wind speed as the wind speed deviation, and statistically analyze the mean and variance of the wind speed deviation under different wind speed and wind direction sectors as the dynamic feature parameter. S322, according to the wind direction sector corresponding to the spatiotemporal registration sample set, the sea-based measured wind speed is grouped and statistically analyzed, the sea-based measured average wind speed under each wind direction sector and the sea-based measured global average wind speed for all wind directions are calculated, and the ratio of the sea-based measured average wind speed to the sea-based measured global average wind speed is used as the wind speed acceleration ratio. S323, according to the time series corresponding to the spatiotemporal registration sample set, extract the time series curve and rate of change of the measured wind speed at the sea base as the continuous wind speed feature of the sea base at a single point.

[0027] In the specific implementation process, from the constructed spatiotemporal registration sample set, the satellite-inverted wind speed and the sea-based measured wind speed in each pair of data are extracted, and the difference between the two (e.g., satellite-inverted wind speed minus sea-based measured wind speed) is calculated as the wind speed deviation for a single observation. The wind speed range in the registration sample set is divided into several continuous wind speed intervals (e.g., with a step size of 1 m / s), and the 360-degree wind direction is divided into several wind direction sectors with equal angles (e.g., 16 azimuth sectors). All registration samples are traversed, and the wind speed deviation is classified into a two-dimensional grid that combines the corresponding wind speed interval and wind direction sector. The arithmetic mean and sample variance of the wind speed deviation contained in each two-dimensional grid are calculated to characterize the central tendency and dispersion of the systematic error under the specific wind speed and wind direction conditions. The set of deviation mean and variance obtained from all grids constitutes the dynamic feature parameter. The wind sector affiliation corresponding to each record in the spatiotemporal registration sample set is obtained, and this is used as a grouping label to classify the sea-based measured wind speeds. For each wind sector, the arithmetic mean of all sea-based measured wind speeds falling within that sector is calculated to obtain the sea-based measured average wind speed under each wind sector. At the same time, without considering wind direction classification, the total arithmetic mean of all sea-based measured wind speeds in the spatiotemporal registration sample set is calculated to obtain the global sea-based measured average wind speed. The sea-based measured average wind speed of each wind sector is divided by the global sea-based measured average wind speed to obtain a dimensionless ratio. This ratio reflects the acceleration or attenuation effect of local wind speed relative to the overall average level under specific wind direction conditions, and is defined as the wind speed acceleration ratio under that wind sector. Using time as the horizontal axis and measured wind speed at sea as the vertical axis, wind speed time-series evolution curves for sea-based wind measurement points are plotted in chronological order. These time-series curves are extracted as the basic characteristics reflecting the dynamic fluctuation process of wind speed. Furthermore, the ratio of the difference in wind speed between adjacent time steps (i.e., the wind speed value at the later moment minus the wind speed value at the previous moment) to the time step is calculated on the time-series curve to obtain the instantaneous rate of change sequence of wind speed over time, which is used to quantify the characteristics of drastic rises and falls and the speed of fluctuations in wind speed. The time-series curves and their corresponding wind speed rate of change sequences are combined to form a continuous wind speed feature at a single sea-based point that can simultaneously characterize both the continuous state and instantaneous dynamic changes of wind speed.

[0028] In the embodiments of this application, step S4 specifically includes: S41, using the sea-based measured wind speed as a reference benchmark, the mean and variance of wind speed deviation in the dynamic characteristic parameters are used as input nodes, and the ratio of satellite-retrieved wind speed to sea-based measured wind speed is used as the target output node to construct a multilayer perceptron. S42, the dynamic feature parameters are input into the multilayer perceptron, and the network weights are iteratively adjusted through the backpropagation algorithm until the loss function converges, thereby completing the construction of the multi-dimensional wind speed correction model to determine the nonlinear mapping relationship; S43, input the dynamic feature parameters of different wind speed magnitudes and wind direction sector combinations into the trained multi-dimensional wind speed correction model, use the wind speed ratio output by the model as the correction coefficient, and arrange them according to the combination relationship between wind speed magnitude and wind direction sector to generate the wind speed correction coefficient matrix.

[0029] In the specific implementation process, the measured wind speed at sea is used as the true reference benchmark, and a multilayer perceptron (MLP) network structure is constructed, which includes an input layer, at least one hidden layer, and an output layer. For the input layer, the mean and variance of the deviations corresponding to each combination of wind speed magnitude and wind direction sector in the aforementioned extracted dynamic feature parameters are used as two-dimensional feature vectors and mapped to two neuron nodes in the input layer. For the output layer, the true ratio of the satellite-inverted wind speed and the measured wind speed at sea under the same combination is calculated and used as the target output node, thereby representing the target ratio that the satellite wind speed needs to be scaled and corrected under the current error distribution state. At the same time, a nonlinear activation function (such as the ReLU function) is set in the hidden layer to give the network the ability to capture complex nonlinear error relationships. Preprocessed and normalized dynamic feature parameter samples are batch-input into a multilayer perceptron. The data undergoes nonlinear layer-by-layer weighted calculation through the input and hidden layers to generate predicted wind speed ratios at the output layer. A mean squared error (MSE) loss function is constructed to calculate the error loss between the model's predicted wind speed ratio and the actual target wind speed ratio. The backpropagation algorithm is used, combined with an optimizer (such as the Adam optimizer), to calculate the gradient of the loss function with respect to the weights and biases of each layer in the network, and the network parameters are iteratively updated along the gradient descent direction. During the iteration process, the loss value on the validation set is monitored. Training is stopped when the loss function decreases to a preset threshold or no longer decreases significantly in consecutive iterations (i.e., convergence), and the current network parameters are saved. This completes the construction of a multi-dimensional wind speed correction model and determines the nonlinear mapping relationship between input error features and output correction ratios. The model iterates through all possible wind speed levels and wind direction sectors in the target sea area, inputting their corresponding dynamic feature parameters (i.e., the mean and variance of each group's deviation) into the trained multi-dimensional wind speed correction model. The model calculates and outputs the corresponding predicted wind speed ratio through forward inference. The predicted wind speed ratio is directly used as the wind speed correction coefficient for that specific wind speed and wind direction combination. Using wind speed level as the row index and wind direction sector as the column index, all calculated correction coefficients are arranged in a matrix according to their corresponding two-dimensional wind speed-wind direction combination relationship, generating a two-dimensional wind speed correction coefficient matrix covering all wind conditions. This allows for rapid and parallel pixel-by-pixel correction of the initial satellite-retrieved wind speed in subsequent steps through table lookup and matching.

[0030] In the embodiments of this application, step S5 specifically includes: S51, Multiply the wind speed correction coefficient matrix with the initial satellite-inverted wind speed field pixel by pixel to correct the spatial distribution deviation; S52, the wind speed acceleration ratio is used as a spatial scaling factor to weight and allocate the continuous wind speed characteristics of the sea-based single point and map them to a two-dimensional grid space. S53, spline interpolation is used to fill in the continuous time dimension of the satellite observation period to generate the fused and corrected wind field data.

[0031] In the specific implementation process, for each pixel in the initial satellite inversion wind speed field, the initial inversion wind speed value corresponding to that pixel and the wind direction data of its location are extracted; based on the wind speed range to which the wind speed value of that pixel falls and the wind direction sector to which the wind direction data belongs, a two-dimensional index lookup table is performed in the pre-constructed wind speed correction coefficient matrix to accurately match the wind speed correction coefficient corresponding to that pixel under the current wind conditions; traversing all pixels in the entire satellite inversion wind speed field, the initial inversion wind speed value of each pixel is multiplied by its matched correction coefficient pixel by pixel, thereby eliminating the systematic spatial distribution deviation caused by geophysical model functions and large-scale environmental factors, and obtaining the satellite wind speed field after spatial deviation correction; The target sea area is divided into two-dimensional grids according to the spatial resolution rules of the initial satellite inversion wind speed field, and the geographical location of the center point of each grid is obtained. Based on the wind direction sector corresponding to each grid point, the wind speed acceleration ratio under that wind direction is extracted as a spatial scaling factor. For the time series curves and rates of change in the continuous wind speed characteristics of the sea-based single point, the scaling factor is used as a weight to multiply and weight them. That is, it is assumed that the dynamic evolution law of the wind field of the offshore grid points is consistent with that of the sea-based single point, but the amplitude is affected by the local topography or sea surface roughness, resulting in acceleration or attenuation effects. In this way, the continuous wind speed characteristics of the single point are extrapolated and extended to the entire two-dimensional grid space, generating an initial framework of a spatial continuous wind field that covers the entire sea area and has continuous time variation characteristics.

[0032] Since satellite remote sensing is a discrete transit observation, there are a large number of observation gaps between two satellite transits. Therefore, the discrete time series of each grid point in the initial framework of the aforementioned continuous spatial wind field at the satellite observation time are extracted. For each spatial grid point, a cubic spline interpolation algorithm is used, with the satellite observation time as the interpolation node. Utilizing the physical characteristic of smooth wind speed changes over time between adjacent nodes, a piecewise cubic polynomial interpolation function is constructed. This interpolation function is used to calculate the wind speed estimate at each time step (e.g., hourly or 10-minutely) during the satellite observation gap period, achieving continuous and seamless filling of missing time-dimension data. Ultimately, this generates deep-sea fused and corrected wind field data that combines high-precision spatial distribution characteristics with continuous temporal evolution.

[0033] Based on the same inventive concept, embodiments of this application also provide a deep-sea wind field inversion and correction system based on the spatiotemporal coupling of satellite and sea-based multi-source data, corresponding to the above-described method; such as Figure 2 As shown, The system includes: The data acquisition module acquires satellite remote sensing images of the target sea area and sea-based measured wind speeds; The initial inversion module preprocesses the satellite remote sensing image to obtain the backscattering coefficient and incident angle, and combines the auxiliary wind direction data with a preset geophysical model function to obtain the initial satellite inversion wind speed field. The feature extraction module uses the sea-based observation spatial coordinates as anchor points to extract the corresponding pixel wind speed sequence from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set. It also extracts dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculates the wind speed acceleration ratio under each wind direction sector, and extracts continuous wind speed features at a single point on the sea. The model building module, based on the dynamic feature parameters and using the sea-based measured wind speed as a reference, constructs a multi-dimensional wind speed correction model, determines the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and generates a wind speed correction coefficient matrix. The wind field fusion module divides the target sea area into grids, uses the wind speed correction coefficient matrix and the wind speed acceleration ratio to extrapolate the continuous wind speed characteristics of the sea-based single point to the grid space, performs time interpolation and spatial correction on the initial satellite-inverted wind speed field, and generates fused and corrected wind field data.

[0034] Based on the same inventive concept, this application also provides an electronic device. The electronic device of this application includes at least one processor and at least one memory electrically connected to each other. The memory is electrically connected to the processor, and the memory stores instructions executable by the at least one processor. These instructions are executed by the at least one processor to enable the at least one processor to perform the deep-sea wind field inversion correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described above.

[0035] It should be noted that the electrical connections between the various units mentioned above do not necessarily represent the connections between lines. Any indirect connection method can be applied to the embodiments of this application as long as it achieves the purpose of this application.

[0036] Based on the same inventive concept, this application also provides a computer storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, it implements the deep-sea wind field inversion correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described above.

[0037] Based on the same inventive concept, this application also provides a computer program product, which is stored in at least one storage medium; the computer program product includes several instructions to cause at least one computer device to execute the deep-sea wind field inversion correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described above.

[0038] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for inverting and correcting deep-sea wind fields based on spatiotemporal coupling of satellite and sea-based multi-source data, characterized in that, Acquire satellite remote sensing images of the target sea area and sea-based measured wind speeds; The satellite remote sensing image is preprocessed to obtain the backscattering coefficient and incident angle. Combined with auxiliary wind direction data, a preset geophysical model function is input to obtain the initial satellite inversion wind speed field. Using the sea-based observation spatial coordinates as anchor points, the corresponding pixel wind speed sequence is extracted from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set. The dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector are extracted from the spatiotemporal registration sample set, the wind speed acceleration ratio under each wind direction sector is calculated, and the continuous wind speed features of a single point on the sea are extracted. Based on the dynamic characteristic parameters and using the sea-based measured wind speed as a reference, a multi-dimensional wind speed correction model is constructed to determine the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and to generate a wind speed correction coefficient matrix. The target sea area is divided into grids. Using the wind speed correction coefficient matrix and the wind speed acceleration ratio, the continuous wind speed characteristics of the sea-based single point are extrapolated to the grid space. The initial satellite-retrieved wind speed field is interpolated in time and spatially corrected to generate fused and corrected wind field data.

2. The method according to claim 1, characterized in that, The preprocessing of the satellite remote sensing image to obtain the backscattering coefficient and incident angle, combined with auxiliary wind direction data and input into a preset geophysical model function, yields the initial satellite inversion wind speed field, specifically including: The backscattering coefficient and the incident angle are obtained by performing radiometric calibration and polarization correction on the satellite remote sensing image; By combining auxiliary wind direction data with the geophysical model function, the initial satellite-retrieved wind speed field is obtained.

3. The method according to claim 1, characterized in that, The process of extracting corresponding pixel wind speed sequences from the initial satellite-retrieved wind speed field using sea-based observation spatial coordinates as anchor points to construct a spatiotemporal registration sample set specifically includes: Using the sea-based observation spatial coordinates as anchor points, wind speed sequences of pixels at the same location in the initial satellite inversion wind speed field are extracted, and these sequences are paired with the sea-based measured wind speeds at the corresponding times to construct the spatiotemporal registration sample set.

4. The method according to claim 3, characterized in that, The process of extracting dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculating the wind speed acceleration ratio under each wind direction sector, and extracting continuous wind speed features at a single point on the sea basis specifically includes: The satellite-retrieved wind speed is extracted from the spatiotemporal registration sample set and the difference between the retrieved wind speed and the sea-based measured wind speed is calculated as the wind speed deviation. The mean and variance of the wind speed deviation under different wind speed and wind direction sectors are statistically analyzed as the dynamic feature parameters. According to the wind direction sector corresponding to the spatiotemporal registration sample set, the sea-based measured wind speed is grouped and statistically analyzed. The sea-based measured average wind speed under each wind direction sector and the sea-based measured global average wind speed for all wind directions are calculated. The ratio of the sea-based measured average wind speed to the sea-based measured global average wind speed is used as the wind speed acceleration ratio. Based on the time series corresponding to the spatiotemporal registration sample set, the time series curve and rate of change of the measured wind speed at the sea base are extracted as the continuous wind speed characteristics of the sea base at a single point.

5. The method according to claim 1, characterized in that, Based on the dynamic characteristic parameters and using the sea-based measured wind speed as a reference, a multi-dimensional wind speed correction model is constructed to determine the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and to generate a wind speed correction coefficient matrix, specifically including: Using the measured wind speed at sea as a reference, the mean and variance of wind speed deviation in the dynamic characteristic parameters are used as input nodes, and the ratio of satellite-retrieved wind speed to measured wind speed at sea is used as the target output node to construct a multilayer perceptron. The dynamic feature parameters are input into the multilayer perceptron, and the network weights are iteratively adjusted through the backpropagation algorithm until the loss function converges, thereby completing the construction of the multidimensional wind speed correction model to determine the nonlinear mapping relationship. The multi-dimensional wind speed correction model is trained by inputting dynamic feature parameters under different combinations of wind speed magnitude and wind direction sector. The wind speed ratio output by the model is used as the correction coefficient, and the wind speed correction coefficient matrix is ​​generated by arranging the wind speed magnitude and wind direction sector according to the combination relationship.

6. The method according to claim 1, characterized in that, The process of dividing the target sea area into grids, using the wind speed correction coefficient matrix and the wind speed acceleration ratio, extrapolating the continuous wind speed characteristics of a single point on the sea base to the grid space, and performing time interpolation and spatial correction on the initial satellite-retrieved wind speed field to generate fused and corrected wind field data specifically includes: The wind speed correction coefficient matrix is ​​multiplied pixel by pixel by the initial satellite-inverted wind speed field to correct spatial distribution deviation; The wind speed acceleration ratio is used as a spatial scaling factor to weight and allocate the continuous wind speed characteristics of the sea-based single point and map them to a two-dimensional grid space; Spline interpolation is used to continuously fill in the time dimension of the satellite observation periods to generate the fused and corrected wind field data.

7. A deep-sea wind field inversion and correction system based on spatiotemporal coupling of satellite and sea-based multi-source data, characterized in that, The system includes: The data acquisition module acquires satellite remote sensing images of the target sea area and sea-based measured wind speeds; The initial inversion module preprocesses the satellite remote sensing image to obtain the backscattering coefficient and incident angle, and combines the auxiliary wind direction data with a preset geophysical model function to obtain the initial satellite inversion wind speed field. The feature extraction module uses the sea-based observation spatial coordinates as anchor points to extract the corresponding pixel wind speed sequence from the initial satellite inversion wind speed field to construct a spatiotemporal registration sample set. It also extracts dynamic feature parameters of wind speed deviation as a function of wind speed magnitude and wind direction sector from the spatiotemporal registration sample set, calculates the wind speed acceleration ratio under each wind direction sector, and extracts continuous wind speed features at a single point on the sea. The model building module, based on the dynamic feature parameters and using the sea-based measured wind speed as a reference, constructs a multi-dimensional wind speed correction model, determines the nonlinear mapping relationship between satellite-retrieved wind speed and sea-based measured wind speed, and generates a wind speed correction coefficient matrix. The wind field fusion module divides the target sea area into grids, uses the wind speed correction coefficient matrix and the wind speed acceleration ratio to extrapolate the continuous wind speed characteristics of the sea-based single point to the grid space, performs time interpolation and spatial correction on the initial satellite-inverted wind speed field, and generates fused and corrected wind field data.

8. An electronic device, characterized in that, Includes at least one processor and at least one memory electrically connected; The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to execute the deep-sea wind field inversion and correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The computer-readable storage medium stores a computer program. When the computer program is executed by the processor, it implements the deep-sea wind field inversion and correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product is stored in at least one storage medium; The computer program product includes several instructions to cause at least one electronic device to execute the deep-sea wind field inversion and correction method based on the spatiotemporal coupling of satellite and sea-based multi-source data as described in any one of claims 1-6.

Images