Short-term prediction method and system for marine environment based on conditional adversarial network

Abstract

The present invention discloses a method and system for short-term prediction of marine environment based on conditional adversarial network, which specifically relates to the field of short-term prediction of marine environment, including S1, obtaining multi-dimensional original data covering the marine physical environment and remote sensing field; S2, performing spatiotemporal alignment and outlier elimination on the multi-dimensional original data, and using a sliding average method to fill in missing values to obtain standardized data; S3, fusing the standardized data into coupling features reflecting the marine dynamic process and constructing spatiotemporal feature coding; S4, constructing a conditional adversarial network model based on the spatiotemporal feature coding, driving the conditional adversarial network model through a loss function to learn the distribution law of real data, generating prediction features that meet spatiotemporal conditions and completing training; the present invention obtains a three-dimensional spatiotemporal feature tensor reflecting the marine dynamic process through multi-source data fusion and coupled feature engineering, thereby improving the data characterization capability and the accuracy of physical mechanism description.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term prediction of marine environment, and more specifically, to a method and system for short-term prediction of marine environment based on conditional adversarial network. Background Art

[0002] In the field of marine environmental monitoring and research, short-term forecasting is crucial for marine resource development, navigation safety, and ecological environmental protection.

[0003] However, the complex ocean environment is characterized by strong nonlinearity, multi-parameter coupling, and dynamic changes in time and space. Traditional prediction methods such as physical models and statistical models face significant challenges. Physical models rely on a large number of prior assumptions and simplified equations, making it difficult to accurately characterize delicate dynamic processes such as thermohaline variability and internal solitary wave propagation. Statistical models are subject to linear assumptions and are unable to capture the nonlinear correlation between ocean current shear and microwave radiation characteristics, resulting in limited prediction accuracy. At the same time, models based on a single data source cannot fully utilize multi-dimensional observational information and lack adaptability to differences in time and space scales. Under extreme conditions, the risk of prediction failure is high.

[0004] Therefore, there is an urgent need for a short-term prediction method and system for the ocean environment based on conditional adversarial networks. By constructing a generative adversarial framework for multi-scale feature fusion, the joint prediction of ocean temperature and salinity, flow field and remote sensing field can be achieved, the bottleneck problem of traditional models in nonlinear modeling and multi-source data fusion can be solved, and the spatiotemporal resolution and accuracy of short-term predictions can be improved. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for short-term prediction of the marine environment based on a conditional adversarial network, and solves the problems raised in the above-mentioned background technology through the following scheme.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for short-term prediction of marine environment based on conditional adversarial network, comprising:

[0007] S1. Acquire multi-dimensional raw data covering the ocean physical environment and remote sensing fields;

[0008] S2. Performing spatiotemporal registration and outlier removal on the multi-dimensional raw data, and filling in missing values using a sliding average method to obtain standardized data;

[0009] S3, fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature codes;

[0010] S4. Constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model to learn the distribution law of real data through a loss function, generating prediction features that meet spatiotemporal conditions and completing training;

[0011] S5. Use the trained model to generate short-term prediction results and quantitatively evaluate the prediction accuracy through independent data sets to verify whether it meets the application requirements.

[0012] Preferably, the multi-dimensional raw data includes: vertical temperature gradient of seawater collected by temperature-salinity-depth instrument and the coefficient of variation of thermocline depth ; Obtain the standard deviation of surface seawater salinity through remote sensing inversion combined with buoy measurement ; Measure the vertical shear intensity Uz of the geostrophic current through an acoustic Doppler current profiler; Monitor the internal solitary wave propagation velocity Vw through a synthetic aperture radar; Collect the microwave brightness temperature polarization difference Td through the AMSR2 microwave radiometer, and obtain the ocean current radial velocity shear SC through the high-frequency ground wave radar network.

[0013] Preferably, the spatiotemporal registration: unify the data of different devices into the WGS84 geographic coordinate system, align the timestamps to the UTC second level, and unify the spatial grid resolution to 1 km × 1 km; the outlier removal: adopt the 3σ principle, that is, under the normal distribution assumption, data exceeding the mean ± 3 times the standard deviation is regarded as abnormal, and obvious error values are eliminated in combination with physical constraints; the sliding average method is used to fill missing values: cubic spline interpolation is used for short time series with ≤ 3 missing sampling points, and the long time series are missing and filled with the mean of historical data for the same period.

[0014] Preferably, the coupling characteristics include a thermohaline dynamic coupling characteristic model and a flow field-remote sensing characteristic fusion model; the thermohaline dynamic coupling characteristic model is used to convert the vertical temperature gradient of seawater in the standardized data into , standard deviation of surface seawater salinity and the coefficient of variation of thermocline depth Imported into the thermohaline dynamic coupling characteristic model, the thermohaline gradient coupling index is obtained ,in, represents the average depth of the thermocline, Represents the reference temperature gradient; the flow field-remote sensing feature fusion model is used to import the microwave brightness temperature polarization difference Td, the geostrophic flow vertical shear intensity Uz and the internal solitary wave propagation speed Vw in the standardized data into the flow field-remote sensing feature fusion model to obtain the microwave polarization flow field response factor ,in, Represents the microwave brightness temperature difference threshold; the construction of spatiotemporal feature coding: the temperature-salinity gradient coupling index CI, the microwave polarization flow field response factor RF and the ocean current radial velocity shear SC are expanded in spatiotemporal dimensions to construct a three-dimensional feature tensor , where T represents the length of the time series, represents the spatial grid dimension, and C=3 represents the number of feature channels.

[0015] Preferably, the conditional adversarial network model includes a generator and a discriminator. The generator adopts a multi-layer spatiotemporal convolutional network as the generator G, and the input includes a conditional vector c and a noise vector z; the generator network structure includes: an input layer, a transposed convolution layer, and a generator output layer; the discriminator D includes hole convolution branches with different expansion rates, which respectively capture local details and regional macro features, specifically: a discriminator input layer, a multi-branch convolution, and a fusion layer.

[0016] Preferably, the input layer: inputs the concatenated conditional vector c and the noise vector, the dimension is dc+dz, where dc represents the conditional vector dimension, which is used to describe the number of features contained in the conditional vector c, and dz represents the noise vector dimension, which is used to describe the number of features contained in the noise vector z; the transposed convolution layer: 3 layers, each layer is followed by a ReLU activation function, and the feature map size is gradually expanded to ; The generator output layer: the convolutional layer generates the predicted feature tensor , the activation function is Tanh; the discriminator input layer: receives the real ocean environment feature tensor F or generates the ocean environment feature tensor f, the dimension is The multi-branch convolution: parallel branches use dilation rates of 1, 3, and 5 to extract multi-scale features; the fusion layer: the features of each branch are fused through global average pooling to output the discrimination probability .

[0017] Preferably, the loss function includes a generator loss function and a discriminator loss function; the generator loss function is specifically expressed as: ; The discriminator loss function is specifically expressed as: ;in, Represents statistical average, F represents the real ocean environment feature tensor, f represents the generated ocean environment feature tensor, pd represents the real data probability distribution, D() represents the discriminator, G() represents the generator, Represents the predicted feature tensor output by the generator, z represents the noise vector that obeys the standard normal distribution, and c represents the conditional vector containing spatiotemporal information. represents the joint probability distribution of noise and conditional, represents the balance coefficient, Represents L1 loss, which is used to measure the absolute error between generated features and true features.

[0018] Preferably, the short-term prediction result is the ocean environment characteristic tensor for the next 12-48 hours. The quantitative evaluation uses the root mean square error, mean absolute error, and correlation coefficient. The root mean square error is specifically expressed as: , the mean absolute error is specifically expressed as: , the correlation coefficient is specifically expressed as: , where N represents the number of validation samples, represents the i-th real ocean environment feature tensor, represents the i-th generated ocean environment feature tensor, Represents the mean value of the real ocean environment characteristic tensor.

[0019] Preferably, the marine environment short-term prediction system based on the conditional adversarial network includes:

[0020] Data acquisition module: covers multi-dimensional raw data of ocean physical environment and remote sensing field;

[0021] Data processing module: performs spatiotemporal registration and outlier removal on the multi-dimensional raw data, and uses a sliding average method to fill in missing values to obtain standardized data;

[0022] Data fusion module: fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature coding;

[0023] Model construction and training module: constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model through the loss function to learn the distribution law of real data, generating prediction features that meet the spatiotemporal conditions and completing training;

[0024] Prediction Evaluation Module: Uses the trained model to generate short-term prediction results, and quantitatively evaluates the prediction accuracy through independent data sets to verify whether it meets application requirements.

[0025] Technical effects and advantages of the present invention:

[0026] 1. This invention obtains a three-dimensional spatiotemporal feature tensor reflecting ocean dynamic processes through multi-source data fusion and coupled feature engineering. This solves the problem that traditional single-data source models cannot capture the multi-parameter coupling laws, and improves the data representation capability and the accuracy of the depiction of physical mechanisms.

[0027] 2. This invention uses a conditional adversarial network generator and a multi-scale discriminator to collaboratively train and obtain ocean environment prediction features that meet spatiotemporal conditions. This solves the problem of traditional physical models' linear assumptions and statistical models' insufficient nonlinear characterization capabilities, and achieves the benefit of adaptively capturing the strong nonlinear dynamics of the ocean environment. Through multi-scale feature fusion and spatiotemporal convolutional architecture, prediction results with high spatiotemporal resolution are obtained, solving the problem of traditional models' insufficient adaptability to different spatial scales and improving the spatiotemporal consistency of prediction results.

[0028] 3. The present invention obtains quantitatively evaluated model prediction accuracy through independent data set verification and multi-dimensional evaluation indicators, which solves the problem of doubtful prediction reliability caused by the lack of an objective verification system in traditional methods, and achieves the benefit of ensuring the practical application value of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of the method structure of the present invention;

[0030] Figure 2 Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] As attached Figure 1 The short-term prediction method of the ocean environment based on the conditional adversarial network shown includes:

[0033] S1. Acquire multi-dimensional raw data covering the ocean physical environment and remote sensing fields;

[0034] In this embodiment, it should be specifically explained that: the multi-dimensional raw data includes: vertical temperature gradient of seawater collected by temperature-salinity-depth instrument and the coefficient of variation of thermocline depth ; Obtain the standard deviation of surface seawater salinity through remote sensing inversion combined with buoy measurement ; Measure the vertical shear intensity Uz of the geostrophic current through an acoustic Doppler current profiler; Monitor the internal solitary wave propagation velocity Vw through a synthetic aperture radar; Collect the microwave brightness temperature polarization difference Td through the AMSR2 microwave radiometer, and obtain the ocean current radial velocity shear SC through the high-frequency ground wave radar network.

[0035] S2. Performing spatiotemporal registration and outlier removal on the multi-dimensional raw data, and filling in missing values using a sliding average method to obtain standardized data;

[0036] In this embodiment, it is specifically necessary to explain that: the spatiotemporal registration: the data of different devices are unified into the WGS84 geographic coordinate system, the timestamps are aligned to the UTC second level, and the spatial grid resolution is unified to 1 km × 1 km; the outlier removal: the 3σ principle is adopted, that is, under the normal distribution assumption, data exceeding the mean ± 3 times the standard deviation is regarded as anomaly, and obvious error values are eliminated in combination with physical constraints; the sliding average method is used to fill missing values: cubic spline interpolation is used for short time series with ≤ 3 sampling points missing, and the long time series missing is filled with the mean of historical data of the same period; it should be understood that the raw data of the marine environment have the following problems: spatiotemporal heterogeneity: the sampling frequency and spatial resolution of different devices are inconsistent, and direct use will lead to feature calculation deviation; noise and outliers: sensor errors, ocean surface white foam interference, etc. may cause the raw data to contain significant outliers; missing values: due to equipment failure or observation conditions, the raw data may have temporal or spatial missing. By generating standardized data, we provide reliable input for feature engineering: Spatiotemporal registration: Unifying multi-source data onto the same spatiotemporal grid ensures strict spatial and temporal alignment of parameters during feature calculation; Outlier removal: Using physical constraints (e.g., the thermocline temperature gradient must be negative) and statistical methods (the 3σ principle) to eliminate erroneous data and avoid the derivation of erroneous features (e.g., distortion of the coupling index calculated based on the standard deviation of anomalous salinity); Missing value filling: Filling missing points through interpolation or the mean of historical data to ensure the continuity of feature calculation (e.g., the coefficient of variation of thermocline depth requires continuous time series data). Preprocessing eliminates interfering factors, ensuring that feature engineering accurately captures the physical nature of the ocean environment and providing a solid data foundation for efficient training of conditional adversarial networks.

[0037] S3, fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature codes;

[0038] In this embodiment, it is specifically necessary to explain that: the temperature-salinity dynamic coupling characteristic model is used to convert the vertical temperature gradient of seawater in the standardized data into , standard deviation of surface seawater salinity and the coefficient of variation of thermocline depth Imported into the thermohaline dynamic coupling characteristic model, the thermohaline gradient coupling index is obtained ,in, represents the average depth of the thermocline, Represents the reference temperature gradient; the flow field-remote sensing feature fusion model is used to import the microwave brightness temperature polarization difference Td, the geostrophic flow vertical shear intensity Uz and the internal solitary wave propagation speed Vw in the standardized data into the flow field-remote sensing feature fusion model to obtain the microwave polarization flow field response factor ,in, Represents the microwave brightness temperature difference threshold; the construction of spatiotemporal feature coding: the temperature-salinity gradient coupling index CI, the microwave polarization flow field response factor RF and the ocean current radial velocity shear SC are expanded in spatiotemporal dimensions to construct a three-dimensional feature tensor , where T represents the length of the time series, represents the spatial grid dimension, and C=3 represents the number of feature channels. What needs to be understood is: The synergistic effect of the enhanced temperature-salinity gradient is that when the two have opposite signs (e.g., warm water has low salinity, cold water has high salinity), the product is negative, indicating strong density stratification; when the signs are the same, the product is positive, which may correspond to a mixed layer or abnormal stratification. The denominator uses square root operation to suppress the dominant role of extreme values, making the index focus more on the relative changes in the gradient rather than the absolute size; The relative strength of the thermocline depth fluctuations is highlighted by nonlinear transformation (squaring + normalization). When , the term value approaches 0, indicating that the stratification is stable; when When , the term value approaches 1, reflecting the violent fluctuation of stratification; The exponential decay function describes the marginal effect of the temperature gradient. When the temperature gradient exceeds the reference value, the decay accelerates. The direction information of the flow field shear is retained. When Uz>0 (the velocity increases with depth), the sign is positive, which may correspond to the upwelling area; when Uz<0, the sign is negative, which may correspond to downwelling or vertical mixing suppressed by the thermocline; The numerator is the product of microwave difference and internal wave velocity, which represents the energy synergy effect; the denominator is the sum of flow field shear and internal wave velocity, which reflects the competitive contribution of the two to sea surface roughness. , the term value approaches |Td|, highlighting the modulation of the microwave signal by the internal wave; when the flow field shear dominates, the term value tends to the flow field energy characteristics; Simulates the nonlinear response of microwave brightness temperature differences. When the difference is less than the threshold, the exponential term approaches 0 and the response factor is suppressed (noise dominates); when the difference exceeds the threshold, the term value approaches 1 and the response factor increases linearly. Constructs a three-dimensional feature tensor The structure retains the spatiotemporal correlation of the data, making it easier for the subsequent spatiotemporal convolutional network to extract dynamic features.

[0039] S4. Constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model to learn the distribution law of real data through a loss function, generating prediction features that meet spatiotemporal conditions and completing training;

[0040] In this embodiment, it is specifically necessary to explain that: the conditional adversarial network model includes a generator and a discriminator. The generator adopts a multi-layer spatiotemporal convolutional network as the generator G, and the input includes a conditional vector c and a noise vector z; the generator network structure includes: an input layer, a transposed convolution layer, and a generator output layer; the discriminator D includes hole convolution branches with different expansion rates, which respectively capture local details and regional macro features, specifically: a discriminator output layer, a multi-branch convolution, and a fusion layer. The input layer: concatenates the conditional vector c and the noise vector z, with a dimension of dc+dz, where dc represents the conditional vector dimension, which is used to describe the number of features contained in the conditional vector c, and dz represents the noise vector dimension, which is used to describe the number of features contained in the noise vector z; the transposed convolution layer: 3 layers, each layer is followed by a ReLU activation function, which gradually expands the feature map size to ; The generator output layer: the convolutional layer generates the predicted feature tensor , the activation function is Tanh; the discriminator output layer: receives the real ocean environment feature tensor F or generates the ocean environment feature tensor f, the dimension is The multi-branch convolution: parallel branches use dilation rates of 1, 3, and 5 to extract multi-scale features; the fusion layer: the features of each branch are fused through global average pooling to output the discrimination probability The loss function includes a generator loss function and a discriminator loss function; the generator loss function is specifically expressed as: ; The discriminator loss function is specifically expressed as: ;in, Represents statistical average, F represents the real ocean environment feature tensor, f represents the generated ocean environment feature tensor, pd represents the real data probability distribution, D() represents the discriminator, G() represents the generator, Represents the predicted feature tensor output by the generator, z represents the noise vector that obeys the standard normal distribution, and c represents the conditional vector containing spatiotemporal information. represents the joint probability distribution of noise and conditional, represents the balance coefficient, represents the L1 loss, which measures the absolute error between the generated features and the true features. It is important to understand that the conditional vector c is the prior information used to guide the generator and typically contains spatiotemporal conditional parameters, such as: timestamp: used to represent the time node of the prediction target (such as the encoded value of time units such as hours and days); spatial coordinates: used to locate the geographic coordinates of the prediction area (such as the normalized values of longitude and latitude); the noise vector z is a random vector following a standard normal distribution (0, 1), which is used to introduce randomness into the generation process, avoid the model generating a single sample, and enhance the diversity of the prediction results; dc is the sum of the dimensions of the above spatiotemporal parameters. For example, if the conditional vector contains "timestamp + longitude + latitude", then dc=3; dz is an artificially set hyperparameter, for example, dz=100 means inputting 100-dimensional random noise; c is used to force the generator to focus on the ocean environment characteristics of a specific spatiotemporal position (such as the temperature and salinity structure of a certain longitude / latitude in the next 24 hours); z is used to capture the subtle uncertainties that are not fully observed in the ocean environment (such as turbulent pulsations, local eddies, and other high-frequency components that are difficult to model), thereby improving the generalization ability of the prediction results; the transposed convolution layer: 3 layers: the first layer: 4×4 convolution kernel, expansion rate 1, output feature map size 12×50×50, activation function ReLU, used to expand spatial resolution; the second layer: 4×4 convolution kernel, expansion rate 2, output feature map size 24×100×100, activation function ReLU, to capture the mesoscale Degree space features; the third layer: 4×4 convolution kernel, expansion rate 1, output feature map size 24×100×100×3, activation function Tanh, normalize the value to [-1, 1] to match the real data distribution; the multi-branch convolution: the parallel branch uses hollow convolution with expansion rates of 1, 3, and 5: branch 1: expansion rate 1, 3×3 convolution kernel, extracting local features (such as the temperature-salinity coupling intensity of a single grid point); branch 2: expansion rate 3, 3×3 convolution kernel, extracting regional features (such as the current shear trend within a 3×3 grid); branch 3: expansion rate 5, 3×3 convolution kernel, extracting macro features (such as the propagation direction of internal waves within a 5×5 grid) the generator loss function: adversarial loss part: goal: force the sample f=G(z, c) output by the generator to be misclassified by the discriminator as real data (i.e. D(f, c) 1) In ocean prediction, the adversarial loss ensures that the generated temperature-salinity gradient, ocean current shear and other characteristics conform to the actual distribution law of the target sea area (such as the statistical characteristics of the shallow thermocline in summer); L1 loss: The goal is to directly constrain the numerical difference between the generated data and the real data (such as the absolute error of the temperature-salinity gradient coupling index); to prevent the generator from ignoring the actual accuracy of the physical quantity; the discriminator loss function: to make the discriminator output the real data =1, generate data output , thereby improving the classification ability of the discriminator and indirectly forcing the generator to generate higher quality samples.

[0041] S5. Use the trained model to generate short-term prediction results and quantitatively evaluate the prediction accuracy through independent data sets to verify whether it meets the application requirements.

[0042] In this embodiment, it should be specifically noted that the short-term prediction result is the ocean environment characteristic tensor for the next 12-48 hours, and the quantitative evaluation uses the root mean square error, mean absolute error, and correlation coefficient. The root mean square error is specifically expressed as: , the mean absolute error is specifically expressed as: , the correlation coefficient is specifically expressed as: , where N represents the number of validation samples, represents the i-th real ocean environment feature tensor, represents the i-th generated ocean environment feature tensor, Represents the mean value of the real ocean environment characteristic tensor. It is important to understand that the root mean square error (RMSE) measures the overall deviation between the predicted and true values, assigning higher weight to larger errors (because squaring amplifies the impact of outliers). It is suitable for testing a model's predictive ability for extreme ocean events (such as internal solitary wave bursts and strong upwelling). If the RMSE increases significantly in extreme samples, it indicates that the model is inadequately modeling nonlinear mutation processes. The RMSE directly reflects the absolute difference between the predicted and true values, making it easier for business personnel to intuitively understand the error magnitude. The mean absolute error (MAE) calculates the average of the absolute differences between the predicted and true values. It is more robust to outliers (avoiding amplification of individual errors by squaring arithmetic) and reflects the level of prediction bias of the model under normal ocean conditions (such as predicting ocean current speed in calm waters). If the MAE and RMSE are close, it indicates that the data distribution is uniform and there are no significant outliers. The correlation coefficient (R²) measures the degree of linear correlation between the predicted and true values. It ranges from 0 to 1. Values closer to 1 indicate greater consistency between the predicted and true data trends. It is suitable for evaluating a model's ability to capture ocean environmental evolution trends (such as seasonal variations in the thermohaline gradient). A high R² indicates that the model can effectively capture the spatiotemporal evolution of physical quantities. These three metrics quantify model performance using the dimensions of "error margin," "mean deviation," and "trend correlation," respectively, to avoid the one-sidedness of a single metric. For example, if the RMSE is high but the R² is high, it indicates that the model is capturing trends but has systematic biases, which can be corrected through calibration. If the R² is low but the MAE is low, it indicates that the predicted values are close to the mean but lack dynamic adaptability, requiring optimization of the spatiotemporal feature extraction mechanism. By comparing the changes in metrics under different training strategies (such as adjusting coupling features or varying the number of network layers), the model structure is optimized in a targeted manner, forming a closed "prediction-verification-tuning" loop to continuously improve the accuracy and reliability of short-term marine environmental forecasts.

[0043] Based on the above scheme and Figure 2 The present invention also provides a short-term marine environment prediction system based on a conditional adversarial network, comprising:

[0044] Data acquisition module: covers multi-dimensional raw data of ocean physical environment and remote sensing field;

[0045] Data processing module: performs spatiotemporal registration and outlier removal on the multi-dimensional raw data, and uses a sliding average method to fill in missing values to obtain standardized data;

[0046] Data fusion module: fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature coding;

[0047] Model construction and training module: constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model through the loss function to learn the distribution law of real data, generating prediction features that meet the spatiotemporal conditions and completing training;

[0048] Prediction Evaluation Module: Uses the trained model to generate short-term prediction results, and quantitatively evaluates the prediction accuracy through independent data sets to verify whether it meets application requirements.

[0049] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0050] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A short-term prediction method for marine environment based on conditional adversarial network, characterized by: include: S1. Acquire multi-dimensional raw data covering the ocean physical environment and remote sensing fields; S2. Performing spatiotemporal registration and outlier removal on the multi-dimensional raw data, and filling in missing values using a sliding average method to obtain standardized data; S3. Fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature coding; the coupling features include a thermohaline dynamic coupling feature model and a flow field-remote sensing feature fusion model; The temperature-haline dynamic coupling characteristic model is used to convert the vertical temperature gradient of seawater in the standardized data into , standard deviation of surface seawater salinity and the coefficient of variation of thermocline depth Imported into the thermohaline dynamic coupling characteristic model, the thermohaline gradient coupling index is obtained ,in, represents the average depth of the thermocline, Represents the reference temperature gradient; the flow field-remote sensing feature fusion model is used to import the microwave brightness temperature polarization difference Td, the geostrophic flow vertical shear intensity Uz and the internal solitary wave propagation speed Vw in the standardized data into the flow field-remote sensing feature fusion model to obtain the microwave polarization flow field response factor ,in, Represents the microwave brightness temperature difference threshold; the construction of spatiotemporal feature coding: the temperature-salinity gradient coupling index CI, the microwave polarization flow field response factor RF and the ocean current radial velocity shear SC are expanded in spatiotemporal dimensions to construct a three-dimensional feature tensor , where T represents the length of the time series, represents the spatial grid dimension, C=3 represents the number of feature channels; S4. Constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model to learn the distribution law of real data through a loss function, generating prediction features that meet spatiotemporal conditions and completing training; S5. Use the trained model to generate short-term prediction results and quantitatively evaluate the prediction accuracy through independent data sets to verify whether it meets the application requirements.

2. The method for short-term marine environment prediction based on conditional adversarial network according to claim 1 is characterized in that: The multi-dimensional raw data includes: vertical temperature gradient of seawater collected by temperature-salinity-depth instrument and the coefficient of variation of thermocline depth ; Obtain the standard deviation of surface seawater salinity through remote sensing inversion combined with buoy measurement ; Measure the vertical shear intensity Uz of the geostrophic current through an acoustic Doppler current profiler; Monitor the internal solitary wave propagation velocity Vw through a synthetic aperture radar; Collect the microwave brightness temperature polarization difference Td through the AMSR2 microwave radiometer, and obtain the ocean current radial velocity shear SC through the high-frequency ground wave radar network.

3. The method for short-term marine environment prediction based on conditional adversarial network according to claim 1 is characterized in that: The spatiotemporal registration method is as follows: the data of different devices are unified into the WGS84 geographic coordinate system, the timestamps are aligned to the UTC second level, and the spatial grid resolution is unified to 1 km × 1 km; the outlier removal method is as follows: the 3σ principle is adopted, that is, under the normal distribution assumption, data exceeding the mean ± 3 times the standard deviation is regarded as anomaly, and obvious error values are eliminated in combination with physical constraints; the sliding average method is used to fill missing values: cubic spline interpolation is used for short time series with ≤ 3 missing sampling points, and the mean of historical data for the same period is used to fill long time series.

4. The method for short-term marine environment prediction based on conditional adversarial network according to claim 1, characterized in that: The conditional adversarial network model includes a generator and a discriminator. The generator adopts a multi-layer spatiotemporal convolutional network as the generator G, and the input includes a conditional vector c and a noise vector z. The generator network structure includes: an input layer, a transposed convolution layer, and a generator output layer. The discriminator D contains hole convolution branches with different expansion rates, which respectively capture local details and regional macro features, specifically: a discriminator input layer, a multi-branch convolution, and a fusion layer.

5. The method for short-term marine environment prediction based on conditional adversarial network according to claim 4 is characterized in that: The input layer: input splicing condition vector c and noise vector, dimension is dc+dz, where dc represents the condition vector dimension, which is used to describe the number of features contained in the condition vector c, and dz represents the noise vector dimension, which is used to describe the number of features contained in the noise vector z; the transposed convolution layer: 3 layers, each layer is followed by a ReLU activation function, and the feature map size is gradually expanded to ; The generator output layer: the convolutional layer generates the predicted feature tensor The activation function is Tanh; the discriminator input layer: receives the real ocean environment feature tensor F or generates the ocean environment feature tensor f, with a dimension of The multi-branch convolution: parallel branches use dilation rates of 1, 3, and 5 to extract multi-scale features; the fusion layer: the features of each branch are fused through global average pooling to output the discrimination probability .

6. The method for short-term marine environment prediction based on conditional adversarial network according to claim 1, characterized in that: The loss function includes a generator loss function and a discriminator loss function; the generator loss function is specifically expressed as: ; The discriminator loss function is specifically expressed as: ;in, Represents statistical average, F represents the real ocean environment feature tensor, f represents the generated ocean environment feature tensor, pd represents the real data probability distribution, D() represents the discriminator, G() represents the generator, Represents the predicted feature tensor output by the generator, z represents the noise vector that obeys the standard normal distribution, and c represents the conditional vector containing spatiotemporal information. represents the joint probability distribution of noise and conditional, represents the balance coefficient, Represents L1 loss, which is used to measure the absolute error between generated features and true features.

7. The method for short-term marine environment prediction based on conditional adversarial network according to claim 1, characterized in that: The short-term prediction result is the ocean environment characteristic tensor for the next 12-48 hours. The quantitative evaluation uses the root mean square error, mean absolute error, and correlation coefficient. The root mean square error is specifically expressed as: , the mean absolute error is specifically expressed as: , the correlation coefficient is specifically expressed as: , where N represents the number of validation samples, represents the i-th real ocean environment feature tensor, represents the i-th generated ocean environment feature tensor, Represents the mean value of the real ocean environment characteristic tensor.

8. A short-term ocean environment prediction system based on a conditional adversarial network, used to implement the short-term ocean environment prediction method based on a conditional adversarial network as described in any one of claims 1 to 7, characterized in that: include: Data acquisition module: covers multi-dimensional raw data of ocean physical environment and remote sensing field; Data processing module: performs spatiotemporal registration and outlier removal on the multi-dimensional raw data, and uses a sliding average method to fill in missing values to obtain standardized data; Data fusion module: fusing the standardized data into coupling features reflecting ocean dynamic processes and constructing spatiotemporal feature coding; Model construction and training module: constructing a conditional adversarial network model based on the spatiotemporal feature encoding, driving the conditional adversarial network model through the loss function to learn the distribution law of real data, generating prediction features that meet the spatiotemporal conditions and completing training; Prediction Evaluation Module: Uses the trained model to generate short-term prediction results, and quantitatively evaluates the prediction accuracy through independent data sets to verify whether it meets application requirements.

Images