Method and system for synchronously estimating sea surface wind vector and sea surface current vector

Abstract

The present invention discloses a method for synchronously estimating sea surface wind vectors and sea surface current vectors using a drifting wave buoy. The method comprises the following steps: obtaining the wave directional spectrum and the longitudinal and latitudinal components of the drift velocity vector measured by an on-site drifting wave buoy; inputting the obtained data into a deep neural network model comprising multiple loop-closure convolution layers and a multi-layer perceptron composed of multiple fully connected layers; the loop-closure convolution layer equates the data at 0° and 360° in the two-dimensional matrix of the wave directional spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel; after feature extraction from the two-dimensional matrix of the wave directional spectrum by the multiple loop-closure convolution layers, the extracted features are vectorized and input into the multi-layer perceptron along with the longitudinal and latitudinal components of the drift velocity vector, which then outputs the sea surface wind vector and the current vector. The present invention achieves synchronous quantitative estimation of sea surface wind vectors and sea surface current vectors based on a drifting wave buoy.

Description

Technical Field

[0001] The present invention relates to the field of oceanography, and in particular to a method and system for synchronously estimating a sea surface wind vector and a sea surface current vector by using a drifting wave buoy. Background Art

[0002] Winds, waves, and currents on the ocean surface are the most significant dynamic phenomena in the ocean. They are the most fundamental dynamic processes affecting momentum, energy, and material exchange at the ocean surface. They are also the fundamental physical quantities of greatest interest in research in disciplines such as oceanography and meteorology. They are also essential parameters of the ocean dynamic environment for nearly all marine engineering and offshore operations. It can be said that surface winds, waves, and currents are closely related to nearly all human activities at sea.

[0003] Currently, simultaneous in-situ observations of sea surface wind, waves, and currents are primarily obtained using integrated hydrometeorological buoys equipped with wind, wave, and current sensors. To observe wind, these buoys require a suitable platform and mast for the meteorological equipment, resulting in a tall and bulky buoy. Furthermore, the installation of current observation equipment and the corresponding anchoring system further increase the size, weight, and cost of the buoy system. The large size also necessitates specialized voyages for deployment and maintenance, further increasing observation costs and contributing to the scarcity of simultaneous in-situ observations of sea surface wind, waves, and currents. Furthermore, wind speed in the wave boundary layer varies significantly with altitude. For wind speed observation, it is crucial to maintain the anemometer at a stable altitude to avoid underestimating wind speeds due to tilting and swaying of the buoy with the wind. Wave-induced swaying of the buoy can also contaminate wind and current measurements with oscillation velocity signals. Therefore, observations of sea surface wind speed and currents require excellent buoy stability. However, wave observation requires that the buoys be able to fully respond to wave motions of varying frequencies and maintain good wave-following performance under varying sea conditions. This conflict between the stability requirements for wind and current measurements and the wave-following requirements for wave measurements prevents these buoys from accurately measuring wind, current, and waves simultaneously.

[0004] In contrast, drifting wave buoys are small, lightweight, and inexpensive, making them easy to deploy and deploy in large numbers, making them suitable for establishing observation networks. Their miniaturized design allows them to follow waves better, enabling more accurate wave measurements than large and medium-sized buoys. However, their miniaturization makes it difficult to install anemometers and current monitoring equipment, making it impossible to directly measure surface wind and current vectors. While installing a sail on a drifting buoy can significantly increase its ability to follow currents, allowing the average surface current velocity at the location it passes through to be calculated based on the buoy's trajectory (using the Lagrangian method), the significant water resistance of the sail affects the buoy's motion characteristics, making it unable to respond to wave motion. Consequently, drifting buoys with sails cannot measure waves. Furthermore, the presence of a sail increases the size, weight, and deployment complexity of the drifting buoy system, significantly increasing the cost of large-scale, operational deployment, especially in open ocean waters. This problem limits the application of drifting wave buoys to a certain extent.

[0005] Ocean waves are directly driven by the surface wind field. In particular, their high-frequency components can be viewed as the integral of local wind effects modulated by low-frequency, long-wave waves. Therefore, it is entirely possible to invert the prevailing wind speed and direction from the wave spectrum measured by wave buoys. Phillips (1985) proposed that, under the forcing of a uniform and stable wind field, gravity waves in the high-frequency portion of the wave spectrum (above 1.3 times the peak frequency of the wind-wave spectrum) will also reach an energy equilibrium state under the combined control of wind input, breakup dissipation, and nonlinear wave-wave interactions. Based on this relationship, Voermans et al. (2020) developed a semi-analytical model for estimating sea surface wind speed and direction using the wave energy spectrum and mean directional Fourier coefficients observed by anchored buoys. However, this model is overly simplified and fails to account for numerous factors, including variations in wave span, low-frequency modulation, differences in wave direction responses at different energy levels, and the inaccuracy of semi-analytical methods in estimating the equilibrium domain. The resulting estimates are extremely low in accuracy and completely fail to meet the needs of observational operations. Jiang (2022) improved this method by taking into account the effects of wave spread changes and high and low frequency modulation, and obtained an estimation model of wind speed and direction through adaptive balance domain estimation of deep neural networks, achieving basically usable accuracy.

[0006] However, both of these models are based on the first two or five Fourier coefficients of the wave spectrum and do not consider the nonlinear interactions between different frequency bands and directions in the directional spectrum. Therefore, there is still significant room for improvement in accuracy. More importantly, both models are based on anchored buoys and are not applicable to drifting buoys. This is because the motion of a drifting buoy is a superposition of the combined effects of wind, waves, and current fields. The presence of the current field will cause the Doppler effect of wave fluctuations, changing the spatial distribution of the wave characteristic lines and affecting the dispersion relationship of the wave fluctuations. Whether the buoy moves with the wind and current will also directly affect the buoy's motion response. As a result, the model based on fixed-point observations (anchored buoys) cannot be directly applied to moving-point observations (drifting buoys). Because the sea surface wind vector cannot be accurately estimated, there is currently no method for estimating the sea surface current vector using drifting wave buoys. Summary of the Invention

[0007] The main purpose of the present invention is to provide a method for calculating the sea surface wind vector and flow vector at a location by using the wave spectrum information and trajectory information measured by a drifting wave buoy.

[0008] The technical solution adopted in the present invention is:

[0009] A method for synchronously estimating a sea surface wind vector and a sea surface current vector using a drifting wave buoy is provided, comprising the following steps:

[0010] S1. Obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information;

[0011] S2. Inputting the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model, where the deep neural network model is a pre-trained model including a multi-layer perceptron consisting of multiple loop convolutional layers and multiple fully connected layers;

[0012] S3, the loop convolution layer equates the data at 0° and 360° in the two-dimensional matrix of the wave direction spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel; after multiple loop convolution layers extract the features of the two-dimensional matrix of the wave direction spectrum, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, which outputs the sea surface wind vector and flow vector.

[0013] Following the above technical solution, during the training process of the deep neural network model, the input data of the training samples are the wave direction spectrum and the longitudinal and latitudinal components of the drift velocity vector observed by the drifting wave buoy within a preset sampling time, and the output data of the training samples are the sea surface wind vector and sea surface current vector data within a certain range of the position of the drifting wave buoy in a similar time period.

[0014] According to the above technical solution, the observation data of the sea surface wind vector used for training is the wind vector data obtained by on-site tracking observation within a certain range of the location of the wave direction spectrum observation data, or the wind vector data obtained by satellite remote sensing observation or reanalysis within a certain range of the location of the wave direction spectrum observation data;

[0015] The observation data of the sea surface current vector used for training is the sea surface current vector data observed by on-site tracking within a certain range of the location of the wave directional spectrum observation data, or the sea surface current vector observation data of ground wave radar within a certain range of the location of the wave directional spectrum observation data.

[0016] Following the above technical solution, each loop convolution layer includes a loop convolution module, a batch normalization module, an activation function module and a pooling module.

[0017] Following the above technical solution, the first loop convolution layer converts the two-dimensional matrix of the wave directional spectrum in the polar coordinate system into a three-dimensional matrix. The dimensions of the two-dimensional matrix are direction and frequency, and the dimensions of the three-dimensional matrix are channel, direction, and frequency.

[0018] Following the above technical solution, when training the deep neural network model, its training goal is to minimize the vector mean square error VMSE;

[0019]

[0020] Where N is the number of training samples, and are the model output in the i-th sample and the standardized sea surface wind vector data in the training sample, δ W_i Indicates whether the stroke vector data of the i-th sample is missing. If not, δ W_i =1, otherwise δ W_i =0; and are the model output in the i-th sample and the standardized surface current vector data in the training sample, δ C_i Indicates whether the flow vector data in the i-th sample is missing. If not, δ W_i =1, otherwise δ W_i =0.

[0021] For the standardized sea surface wind vector and sea surface current vector data output by the model, an inverse normalization transformation is performed to obtain the actual output sea surface wind vector and sea surface current vector.

[0022] Following the above technical solution, if the directional resolution of the wave directional spectrum is N° (N is a divisor of 360), there are 360 / N directional grids in the wave directional spectrum.

[0023] The present invention also provides a system for synchronously estimating sea surface wind vectors and sea surface current vectors using a drifting wave buoy, comprising:

[0024] An input data acquisition module is used to obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information;

[0025] The estimation module is used to input the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model. The deep neural network model is a pre-trained model, including a multi-layer perceptron composed of multiple loop convolution layers and multiple fully connected layers. The loop convolution layer equates the data in the 0° and 360° directions in the two-dimensional matrix of the wave directional spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel. After the two-dimensional matrix of the wave directional spectrum is extracted through multiple loop convolution layers, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, and the sea surface wind vector and flow vector are output through the multi-layer perceptron.

[0026] Following the above technical solution, each loop convolution layer in the deep neural network model includes a loop convolution module, a batch normalization module, an activation function module and a pooling module.

[0027] The present invention also provides a computer storage medium storing a computer program executable by a processor, wherein the computer program executes the method for synchronously estimating the sea surface wind vector and the sea surface current vector using a drifting wave buoy as described in the above technical solution.

[0028] The beneficial effect of the present invention is as follows: the present invention adopts a data-driven approach to the complex nonlinear relationship in the process of mapping the wave direction spectrum and the drift velocity vector of the buoy to the sea surface wind vector / current vector, and uses a trained deep learning model to quantify this mapping relationship, ultimately achieving synchronous quantitative estimation of the sea surface wind vector and the sea surface current vector.

[0029] During the deep learning modeling process, a polar coordinate loop convolution layer was designed to process the wave directional spectrum, taking into account the equivalence of the 0° and 360° directions of the two-dimensional data features of the wave directional spectrum. After compressing the features, the layers were concatenated with the buoy drift velocity vector and put into a fully connected multi-layer perceptron for processing. This solved the problem of spectral edge direction discontinuity caused by the direct convolution of the wave directional spectrum.

[0030] Furthermore, the present invention realizes the indirect measurement of the sea surface wind vector and the sea surface current vector by using the observation data of the drifting wave buoy. While retaining the advantages of the wave buoy such as small size, light weight, low cost, easy deployment, and more accurate measured wave spectrum, it solves the problem that the drifting wave buoy cannot obtain the sea surface wind vector and current vector due to its miniaturization. In a true sense, it realizes the joint synchronous observation of sea surface wind, waves and current based on the drifting wave surface wind vector and sea surface current vector wave buoys, greatly improving the application space and application scenarios of the drifting wave buoy, and greatly reducing the cost of obtaining the wave spectrum, sea surface wind vector and current vector data. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 is a flow chart of a method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to an embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the wave direction spectrum;

[0034] Figure 3 Schematic diagram of the deep neural network model structure of an embodiment of the present invention;

[0035] Figure 4 Schematic diagram comparing the loop convolution structure used in the deep neural network model of the present invention with ordinary convolution;

[0036] Figure 5 This is a flowchart of building and training a deep neural network model according to an embodiment of the present invention;

[0037] Figure 6 1 is a schematic diagram of the system structure for synchronously estimating the sea surface wind vector and the sea surface current vector using a drifting wave buoy according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] Example 1

[0040] like Figure 1 As shown, the method for synchronously estimating the sea surface wind vector and the sea surface current vector using a drifting wave buoy according to an embodiment of the present invention includes the following steps:

[0041] S1. Obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information;

[0042] S2. Inputting the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model, where the deep neural network model is a pre-trained model including a multi-layer perceptron consisting of multiple loop convolutional layers and multiple fully connected layers;

[0043] S3, the loop convolution layer equates the data at 0° and 360° in the two-dimensional matrix of the wave direction spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel; after multiple loop convolution layers extract the features of the two-dimensional matrix of the wave direction spectrum, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, which outputs the sea surface wind vector and flow vector.

[0044] This embodiment fully considers the modulation of the wave directional spectrum by the interaction of high and low frequencies of waves, as well as the wave dispersion relationship under the combined effects of wind, waves, and currents and the buoy observation process. It estimates the sea surface wind vector using the complete wave directional spectrum and the buoy's drift velocity vector as input.

[0045] At the same time, the drift of a small drifting wave buoy can be regarded as a nonlinear superposition of wind-induced drift, wave-induced drift and current-induced drift, and the drift parameter characteristics of wind drift, wave drift and current drift of standardized products can be regarded as constants. Therefore, on the premise of obtaining the wind vector and the buoy drift velocity vector, the sea surface current vector can be inferred.

[0046] Furthermore, since the sea surface wind vector can be regarded as a function of the directional spectrum and the buoy drift velocity vector, the sea surface current vector can also be synchronously estimated through the directional spectrum, the buoy drift velocity vector and the wind vector.

[0047] In summary, the present invention uses the wave spectrum information and trajectory information measured by the drifting wave buoy and a pre-trained deep neural network model to simultaneously infer the sea surface wind vector and flow vector at the location.

[0048] Example 2

[0049] This embodiment is based on Example 1, with reference to Figure 5 , the construction and training of the deep neural network model in this embodiment mainly includes the following steps:

[0050] Step S101. Obtain the wave direction spectrum measured by the drifting wave buoy (such as Figure 2 The wave directional spectrum information can be obtained from the motion signal of the buoy using a mature wave directional spectrum reconstruction method;

[0051] Step S102: Acquire the trajectory information of the drifting wave buoy during the sampling period (usually 10 to 20 minutes) of the wave directional spectrum measured by the drifting wave buoy in S101, and estimate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information;

[0052] Among them, the longitudinal and latitudinal components of the drift velocity vector V x and V Y They can be obtained by linear least square fitting of the longitudinal and latitudinal positions X and Y in the buoy trajectory versus time t:

[0053] X=V X t+X0,Y=V Y t+Y0

[0054] Step S103. Acquire sea surface wind vector and sea surface current vector data that are close to the temporal and spatial positions of the wave direction spectrum and drift velocity vector observed by the drifting wave buoy in S101 and S102;

[0055] Among them, the data in steps S101, S102, and S103 should be basically consistent in time and space, that is, the two measurements should be no more than a certain distance apart (generally 20 km near the coast and 100 km offshore), and the difference should be no more than several hours (generally 0.5 h near the coast and 3 h offshore). The observation data of the sea surface wind vector can use the wind vector data of on-site tracking observations near the location of the wave directional spectrum observation data in step S101, or the wind vector data of satellite remote sensing observations or reanalysis near the location of the wave directional spectrum observation data in step S101. The observation data of the sea surface current vector can use the sea surface current vector data of on-site tracking observations near the location of the wave directional spectrum observation data in step S101, or the sea surface current vector observation data of ground wave radar near the location of the wave directional spectrum observation data in step S101.

[0056] Step S104: Establish a deep neural network model, the input of the model is the wave direction spectrum and drift velocity vector observed by the drifting wave buoy, and the output of the model is the sea surface wind vector and flow vector.

[0057] The reason for choosing to use deep neural network modeling and selecting the wave direction spectrum and drift velocity vector observed by the drifting wave buoy as the input of the model and the sea surface wind vector as the output is that: the sea surface wind field can simultaneously affect the shape of the wave direction spectrum and the drift velocity of the buoy; the existence of the flow field will cause the Doppler effect of wave fluctuations, change the spatial distribution of the wave characteristic line, affect the dispersion relationship of the fluctuation, and thus affect the shape of the wave direction spectrum; and the movement of the drifting buoy is a nonlinear superposition under the combined action of wind, waves, and flow field. Therefore, in addition to the influence of the flow field, the wave spectrum measured by the buoy will be further nonlinearly modulated by the sea surface wind field. Each of the above relationships is nonlinear and has nested correlations with each other, making it difficult to establish an analytical model to accurately calculate them. However, these principles show that there is a complex quantitative constraint relationship between the wave direction spectrum, drift velocity vector, sea surface wind vector, and sea surface flow vector observed by the drifting wave buoy. Therefore, the present invention uses the deep learning model to model the mining ability of nonlinear quantitative relationships.

[0058] The established deep neural network model structure includes a polar coordinate loop closure convolutional neural network (including multiple layers of loop closure convolutional layers) designed for the wave directional spectrum data structure and a multi-layer perceptron (including multiple fully connected layers), such as Figure 3 As shown in the figure, the model input consists of two parts: the wave directional spectrum, which can be viewed as a two-dimensional matrix in a polar coordinate system, and the X and Y components of the drift velocity vector (corresponding to the longitudinal and latitudinal components). Given the polar coordinate grid characteristics of the wave directional spectrum input, a loopback convolution layer is designed in polar coordinates to extract the directional spectrum features.

[0059] The directional resolution of the wave directional spectrum is N°, so there are 360 / N directional grids in the wave directional spectrum, where N is a divisor of 360. The schematic diagram of the loop convolution is as follows Figure 4As shown: The figure shows a schematic diagram of a two-dimensional matrix of the wave directional spectrum. The corresponding two dimensions are frequency and direction. In this example, the directional resolution of the wave directional spectrum is 10° (a total of 36 directional grids). Convolution kernel n×n, taking the 5×5 convolution kernel as an example, the traditional ordinary convolution method (light gray) needs to further fill two layers of zeros on the left edge of the grid when convolving the "1" grid near 0° to maintain the unchanged matrix size before and after the convolution; and the loop convolution (dark gray) proposed in the present invention, when convolving the "2" grid near 0° (here near refers to the (n-1) / 2 directional grids on both sides with a certain direction grid as the center, and n is the size of the convolution kernel), the data of 340° and 350° on the other side of the matrix are included in the convolution kernel (that is, the convolution is the data of 0°-20° and 340°-350°), so that the data continuity and equivalence between 0° and 350° in the directional space can be fully considered, and the contributions of the grids near 350° and 0° are considered respectively when convolving the grids near 0° and 350°, thereby solving the problem of spectral edge directional discontinuity caused by the direct convolution of the wave directional spectrum.

[0060] After fully extracting the wave directional spectrum through several loop-closure convolutional layers, the output is vectorized and fed into a fully connected multilayer perceptron along with the X and Y components of the drift velocity vector. The output of the multilayer perceptron is the surface wind vector and the flow vector. Each loop-closure convolutional layer in the deep neural network model contains a corresponding loop-closure convolution module, batch normalization module, activation function module, and pooling module.

[0061] Step S105. Using the wave directional spectrum measured by the drifting wave buoy in step S101 and the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy in step S102 as inputs, and the sea surface wind vector data and sea surface current vector data in step S103 as output targets of the model;

[0062] Step S106: Training the deep neural network model in step S104 to obtain a model for synchronously estimating sea surface wind vectors and flow vectors using drifting wave buoys;

[0063] The goal of deep neural network model training is to minimize the following vector mean square error (VMSE):

[0064]

[0065] Where N is the number of training samples, and are the model output in the i-th sample and the standardized sea surface wind vector data in the training sample, δ W_i Indicates whether the stroke vector data of the i-th sample is missing. If not, δ W_i=1, otherwise δ W_i =0; and are the model output in the i-th sample and the standardized surface current vector data in the training sample, δ C_i Indicates whether the flow vector data in the i-th sample is missing. If not, δ W_i =1, otherwise δ W_i = 0. The purpose of standardizing the wind vector and flow vector data is to make the shape of the output space uniform and improve the convergence of the model. W_i and δ C_i Because some training samples may contain missing sea surface wind vectors or current vectors, the introduction of this switch variable allows the model to continue training even when there are some missing wind vectors or current vectors (the label of the missing data needs to be set to an arbitrary value). For the standardized sea surface wind vector and sea surface current vector data output by the model, an inverse normalization transformation is performed to obtain the actual output sea surface wind vector and sea surface current vector.

[0066] Among them, the standardization method can be adopted as follows, and for a given training sample, the standardized wind vector and flow vector can be expressed as:

[0067]

[0068] Where mean and std represent the operations of calculating the mean and standard deviation of the entire training sample, respectively, and ori represents the original data in the training sample. Accordingly, the inverse normalization step can be expressed as:

[0069]

[0070]

[0071] In the formula and They are the normalized wind vector and normalized flow vector output by the model, and the final and are the final estimated wind vector and flow vector, respectively.

[0072] In the absence of surface wind vector and surface current vector data, the wave direction spectrum measured by the drifting wave buoy and the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy are used as the input of the trained model in step S106, and the output obtained by the model is the on-site wind vector and current vector.

[0073] This embodiment uses a data-driven approach to quantify the complex nonlinear relationship between the wave directional spectrum and the buoy's drift velocity vector in mapping to the surface wind vector / current vector. This method utilizes a deep learning model to train a data-driven approach to quantify this mapping relationship. During the deep learning modeling process, a polar coordinate loop convolution module was designed to process the wave directional spectrum based on the two-dimensional data features of the wave directional spectrum and the equivalence of the 0° and 360° directions. After compressing the features, the model is concatenated with the buoy's drift velocity vector and processed again in a fully connected multi-layer perceptron.

[0074] Example 3

[0075] This embodiment is a system for implementing the above method embodiment. Figure 6 As shown, the system of this embodiment using a drifting wave buoy to synchronously estimate the sea surface wind vector and the sea surface current vector includes:

[0076] An input data acquisition module is used to obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information;

[0077] The estimation module is used to input the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model. The deep neural network model is a pre-trained model, including a multi-layer perceptron composed of multiple loop convolution layers and multiple fully connected layers. The loop convolution layer equates the data in the 0° and 360° directions in the two-dimensional matrix of the wave directional spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel. After the two-dimensional matrix of the wave directional spectrum is extracted through multiple loop convolution layers, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, and the sea surface wind vector and flow vector are output through the multi-layer perceptron.

[0078] Among them, each loop convolution layer in the deep neural network model contains a loop convolution module, a batch normalization module, an activation function module and a pooling module.

[0079] The specific functional modules of the system are used to implement each step of the above method embodiment and will not be described in detail here.

[0080] Example 4

[0081] The present application also provides a computer-readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic storage device, a disk, an optical disk, a server, an App store, etc., storing a computer program that, when executed by a processor, implements a corresponding function. When executed by a processor, the computer-readable storage medium of this embodiment implements the method of synchronously estimating the sea surface wind vector and the sea surface current vector using a drifting wave buoy according to the method embodiment.

[0082] In summary, the present invention realizes the indirect measurement of the sea surface wind vector and the sea surface current vector by using the observation data of the drifting wave buoy. While retaining the advantages of the wave buoy such as small size, light weight, low cost, easy deployment, and more accurate measured wave spectrum, it solves the problem that the drifting wave buoy cannot obtain the sea surface wind vector and current vector due to its miniaturization. In a true sense, it realizes the joint synchronous observation of sea surface wind, waves and currents based on the drifting wave surface wind vector and sea surface current vector wave buoys, greatly improving the application space and application scenarios of the drifting wave buoy, and greatly reducing the cost of obtaining the wave spectrum, sea surface wind vector and current vector data.

[0083] Compared with the existing sea surface wind speed and direction estimation method based on the Fourier coefficient of the wave spectrum, the advantages of the present invention are as follows:

[0084] 1. Voermans et al. (2020) developed a semi-analytical model for estimating sea surface wind speed and direction based on the wave energy spectrum and mean directional Fourier coefficients observed by anchored buoys. Because these models failed to account for wave span variations, low-frequency modulation, differences in wave directional response to different energies, nonlinear interactions across different frequency bands and directions within the directional spectrum, and the disruption of fixed-point wave dispersion relations by the drifting buoy's wind and current motion, the resulting model was poorly accurate and exhibited numerous outliers, making it unsuitable for operational applications and unsuitable for drifting wave buoys. Compared to the semi-analytical model of Voermans et al. (2020) that estimates sea surface wind speed and direction based on the wave energy spectrum and average directional Fourier coefficients observed by anchored buoys, the present invention takes into account wave span changes, low-frequency modulation, differences in wave direction responses of different energies, as well as nonlinear interactions between different frequency bands and directions in the entire directional spectrum, and the destruction of the fixed-point wave dispersion relationship by the drifting buoy's wind and current motion. The model is adaptively established through a deep neural network. Therefore, the wind vector estimation model established by the present invention for wave drifting buoys is much more accurate than the model of Voermans et al. (2020), with the root mean square error of wind speed and wind direction both less than 50% of the model of Voermans et al. (2020), and meets the requirements of operational observation for wind vector accuracy.

[0085] 2. Compared with the fully connected deep neural network model established by Jiang (2022) based on the first five Fourier coefficients of the wave spectrum observed by the anchored buoy, the present invention, on the one hand, replaces the Fourier coefficients with the complete wave directional spectrum, taking into account the nonlinear weight effects and interaction effects of different frequency bands and different directions in the complete directional spectrum, and on the other hand, takes into account the destruction of the fixed-point wave dispersion relationship caused by the drifting buoy's wind and current motion by the drift velocity vector information of the buoy as the input of the model. At the same time, in terms of deep neural network design, the present invention introduces a loop convolution module for the input of the complete wave directional spectrum. Through the weight sharing mechanism of the convolutional neural network, the number of free parameters of the model and the number of samples required for training are reduced. Based on the same drifting buoy training samples, the root mean square error of wind speed and wind direction of the wind vector estimation method established by the present invention is only about 70% of that of the Jiang (2022) method, which is more suitable for operational observation.

[0086] 3. More importantly, compared to Voermans et al. (2020) and Jiang (2022), the present invention can provide an accurate estimate of the surface current vector simultaneously with the surface wind vector. Prior to this invention, no relevant technology could estimate the ocean current at the location of a small drifting wave buoy without a sail (since sails would pull on the buoy, affecting its wave-following ability, all drifting wave buoys cannot be equipped with sails).

[0087] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0088] The size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0089] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy, characterized in that: The following steps are involved: S1. Obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information; S2. Inputting the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model, where the deep neural network model is a pre-trained model including a multi-layer perceptron consisting of multiple loop convolutional layers and multiple fully connected layers; S3, the loop convolution layer equates the data at 0° and 360° in the two-dimensional matrix of the wave direction spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel; after multiple loop convolution layers extract the features of the two-dimensional matrix of the wave direction spectrum, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, which outputs the sea surface wind vector and flow vector.

2. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: During the training process of the deep neural network model, the input data of the training samples are the wave direction spectrum and the longitudinal and latitudinal components of the drift velocity vector observed by the drifting wave buoy within the preset sampling time, and the output data of the training samples are the sea surface wind vector and sea surface current vector data within a certain range of the position of the drifting wave buoy in a similar time period.

3. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: The sea surface wind vector observation data used for training is wind vector data obtained by on-site tracking observation within a certain range of the location of the wave direction spectrum observation data, or wind vector data obtained by satellite remote sensing observation or reanalysis within a certain range of the location of the wave direction spectrum observation data; The observation data of the sea surface current vector used for training is the sea surface current vector data observed by on-site tracking within a certain range of the location of the wave directional spectrum observation data, or the sea surface current vector observation data of ground wave radar within a certain range of the location of the wave directional spectrum observation data.

4. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: Each loop convolution layer contains a loop convolution module, a batch normalization module, an activation function module and a pooling module.

5. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: The first loop convolution layer converts the two-dimensional matrix of the wave directional spectrum in the polar coordinate system into a three-dimensional matrix. The dimensions of the two-dimensional matrix are direction and frequency, and the dimensions of the three-dimensional matrix are channel, direction, and frequency.

6. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: When training a deep neural network model, the training goal is to minimize the following vector root mean square error VMSE; Where N is the number of training samples, and are the model output in the i-th sample and the standardized sea surface wind vector data in the training sample, δ W_i Indicates whether the stroke vector data of the i-th sample is missing. If not, δ W_i =1, otherwise δ W_i =0; and are the model output in the i-th sample and the standardized surface current vector data in the training sample, δ C_i Indicates whether the flow vector data in the i-th sample is missing. If not, δ W_i =1, otherwise δ W_i =0; For the standardized sea surface wind vector and sea surface current vector data output by the model, an inverse normalization transformation is performed to obtain the actual output sea surface wind vector and sea surface current vector.

7. The method for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 1, characterized in that: The directional resolution of the wave directional spectrum is N°, so there are 360 / N directional grids in the wave directional spectrum, where N is a divisor of 360.

8. A system for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy, characterized in that: include: An input data acquisition module is used to obtain the wave direction spectrum measured by the on-site drifting wave buoy, and obtain the trajectory information of the buoy drift during its sampling period, and calculate the longitudinal and latitudinal components of the drift velocity vector of the drifting wave buoy based on the trajectory information; The estimation module is used to input the longitudinal and latitudinal components of the wave directional spectrum and the drift velocity vector into a deep neural network model. The deep neural network model is a pre-trained model, including a multi-layer perceptron composed of multiple loop convolution layers and multiple fully connected layers. The loop convolution layer equates the data in the 0° and 360° directions in the two-dimensional matrix of the wave directional spectrum, and simultaneously convolves the data at equal angles on both sides of 0° according to the size of the convolution kernel. After the two-dimensional matrix of the wave directional spectrum is extracted through multiple loop convolution layers, the extracted features are vectorized and input into the multi-layer perceptron together with the longitudinal and latitudinal components of the drift velocity vector, and the sea surface wind vector and flow vector are output through the multi-layer perceptron.

9. The system for synchronously estimating sea surface wind vector and sea surface current vector using a drifting wave buoy according to claim 8, characterized in that: Each loop convolution layer in the deep neural network model contains a loop convolution module, a batch normalization module, an activation function module and a pooling module.

10. A computer storage medium, characterized in that A computer program executable by a processor is stored therein, and the computer program executes the method for synchronously estimating the sea surface wind vector and the sea surface current vector using a drifting wave buoy as described in any one of claims 1 to 7.

Images