Sea condition prediction method, device, equipment, medium and program product

Abstract

The invention belongs to the technical field of sea condition prediction, and particularly relates to a sea condition prediction method, device and equipment, a medium and a program product. The sea condition prediction method comprises the following steps: acquiring sea wave features and wind field features of a target sea area within a preset duration range; respectively extracting sea wave time sequence characteristics and wind field time sequence characteristics from the sea wave characteristics and the wind field characteristics in a preset time length range by utilizing a bidirectional gating circulation unit; respectively determining sea wave space-time attention characteristics and wind field space-time attention characteristics according to the sea wave time sequence characteristics and the wind field time sequence characteristics; according to the sea wave space-time attention features and the wind field space-time attention features, wind wave fusion features are determined; extracting an extreme sensing feature from the wind wave fusion feature by using an extreme sensing module; and predicting the sea condition of the target sea area according to the extreme perception features. And integration of sea condition trend prediction and extreme state intelligent early warning is realized.

Description

Technical Field

[0001] This invention belongs to the technical field of sea state prediction, and specifically relates to a method, apparatus, equipment, medium, and program product for predicting sea states. Background Technology

[0002] High-precision acquisition of wave and sea wind elements is a core aspect of understanding marine environmental conditions during the construction and operation of offshore wind power. These elements not only directly affect the selection of construction windows and wave and wind load analysis of wind turbine foundations and tower structures, but are also closely related to the power output efficiency of wind turbine units, fatigue damage assessment, and platform service life prediction.

[0003] Offshore wind farms operate in complex and variable environments. Extreme wind and wave events (such as sudden strong winds, giant waves, and typhoon aftershocks) often exert a strong impact on wind turbine towers, foundations, submarine cables, and maintenance platforms within a short period. Real-time prediction and early warning of extreme sea conditions are crucial technical aspects for ensuring the safety of wind power equipment and reducing unplanned downtime. Traditional numerical models, such as the Shallow Wave Model (SWAN) and Weather Research and Forecasting (WRF) models, while physically rigorous, rely on external background fields and struggle to quickly respond to locally abrupt wind and wave dynamic conditions. They also suffer from long computation times and difficulty in real-time updates. Statistical regression models and neural network regression models (such as LSTM and GRU), while capable of handling some time-series features, tend to underestimate extreme values ​​when extreme samples are scarce and wind and wave nonlinear coupling is significant, leading to warning failures or delays. Summary of the Invention

[0004] This invention provides a method for predicting sea conditions, comprising: acquiring wave and wind field characteristics of a target sea area within a preset time range; extracting wave and wind field temporal features from the wave and wind field characteristics within the preset time range using a bidirectional gated loop unit; determining wave and wind field spatiotemporal attention features based on the wave and wind field temporal features; determining wind-wave fusion features based on the wave and wind field spatiotemporal attention features; extracting extreme perception features from the wind-wave fusion features using an extreme perception module; and predicting the sea conditions of the target sea area based on the extreme perception features. This method solves the problem in existing technologies where extreme samples are scarce and wind-wave nonlinear coupling is significant, easily leading to underestimation of extreme values ​​and resulting in warning failure or delay. It integrates sea condition trend prediction and intelligent early warning for extreme conditions. Compared with traditional prediction methods based on statistics or single-source measurements, this application significantly improves the accuracy of extreme sea condition identification, prediction timeliness, and deployability, and is applicable to real-time operation monitoring and risk control systems for offshore wind farms.

[0005] To address the aforementioned technical problems, this application proposes five aspects.

[0006] In a first aspect, this application provides a method for predicting sea conditions, comprising: acquiring wave characteristics and wind field characteristics of a target sea area within a preset time range; extracting wave temporal characteristics and wind field temporal characteristics from the wave characteristics and wind field characteristics within the preset time range using a bidirectional gated loop unit; determining wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics based on the wave temporal characteristics and wind field temporal characteristics; determining wind-wave fusion characteristics based on the wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics; extracting extreme perception characteristics from the wind-wave fusion characteristics using an extreme perception module; and predicting the sea conditions of the target sea area based on the extreme perception characteristics. In some embodiments, acquiring the wave characteristics and wind field characteristics of the target sea area includes: acquiring the sea surface echo spectrum of the target sea area within a preset time range using a microwave radar; determining the wave characteristics of the target sea area at each moment within the preset time range based on the sea surface echo spectrum; acquiring the wind speed field at multiple radial points in front of the wind turbine rotor within a preset time range using a laser wind-measuring radar; and determining the wind field characteristics of the target sea area at each moment within the preset time range based on the wind speed field.

[0007] In some embodiments, the step of extracting wave time-series features and wind field time-series features from the wave features and wind field features within a preset time range using a bidirectional gated loop unit includes: time-aligning the wave features and wind field features; inputting the time-aligned wave features and wind field features into the bidirectional gated loop unit; and obtaining the wave time-series features and wind field time-series features through the bidirectional gated loop unit.

[0008] In some embodiments, determining the spatiotemporal attention features of ocean waves and the spatiotemporal attention features of wind fields based on the ocean wave temporal features and the wind field temporal features respectively includes: calculating the ocean wave attention weight of the ocean wave temporal features at each time moment; determining the spatiotemporal attention features of ocean waves based on the ocean wave temporal features and the ocean wave attention weight at each time moment; calculating the wind field attention weight of the wind field temporal features at each time moment; and determining the spatiotemporal attention features of wind fields based on the wind field temporal features and the wind field attention weight at each time moment.

[0009] In some embodiments, extracting extreme sensing features from the wind and wave fusion features using the extreme sensing module includes: calculating the extreme probability score of the wind and wave fusion features using a multilayer perceptron network; determining extreme weights based on the extreme probability score; amplifying the saliency features of the wind and wave fusion features using a nonlinear amplification function to obtain fusion amplification features; and determining the extreme sensing features based on the fusion amplification features, the wind and wave fusion features, and the extreme weights.

[0010] In some embodiments, predicting the sea conditions of the target sea area based on the extreme sensing features includes: using a two-layer fully connected network to predict wave parameters of the target sea area in future time periods based on the extreme sensing features; using a single-layer fully connected network and an activation function to predict the probability of extreme events based on the extreme sensing features; obtaining a preset probability threshold; and determining whether to issue an extreme event warning based on the extreme event probability and the probability threshold.

[0011] Secondly, this application proposes a sea state prediction device, comprising: a first acquisition module for acquiring wave characteristics and wind field characteristics of a target sea area within a preset time range; a first execution module for extracting wave temporal characteristics and wind field temporal characteristics from the wave characteristics and wind field characteristics within the preset time range using a bidirectional gated loop unit; a first determination module for determining wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics based on the wave temporal characteristics and wind field temporal characteristics; a second determination module for determining wind-wave fusion characteristics based on the wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics; a second execution module for extracting extreme perception characteristics from the wind-wave fusion characteristics using an extreme perception module; and a first prediction module for predicting the sea state of the target sea area based on the extreme perception characteristics.

[0012] Thirdly, this application proposes a computer electronic production apparatus, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any of the methods described in the first aspect.

[0013] Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first aspects of the claims.

[0014] Fifthly, this application proposes a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0015] This invention provides a method for predicting sea conditions, comprising: acquiring wave and wind field characteristics of a target sea area within a preset time range; extracting wave and wind field temporal features from the wave and wind field characteristics within the preset time range using a bidirectional gated loop unit; determining wave and wind field spatiotemporal attention features based on the wave and wind field temporal features; determining wind-wave fusion features based on the wave and wind field spatiotemporal attention features; extracting extreme perception features from the wind-wave fusion features using an extreme perception module; and predicting the sea conditions of the target sea area based on the extreme perception features. This method solves the problem in existing technologies where extreme samples are scarce and wind-wave nonlinear coupling is significant, easily leading to underestimation of extreme values ​​and resulting in warning failure or delay. It integrates sea condition trend prediction and intelligent early warning for extreme conditions. Compared with traditional prediction methods based on statistics or single-source measurements, this application significantly improves the accuracy of extreme sea condition identification, prediction timeliness, and deployability, and is applicable to real-time operation monitoring and risk control systems for offshore wind farms. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0017] Figure 1 The main flowchart of a sea state prediction method provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a deep learning model provided in an embodiment of this application; Figure 3 A main structural block diagram of a sea state prediction device provided in an embodiment of this application; Figure 4 This is a structural block diagram of a computer electronic production equipment provided in an embodiment of this application. Detailed Implementation

[0018] High-precision acquisition of wave and sea wind elements is a core aspect of understanding marine environmental conditions during the construction and operation of offshore wind power. These elements not only directly affect the selection of construction windows and wave and wind load analysis of wind turbine foundations and tower structures, but are also closely related to the power output efficiency of wind turbine units, fatigue damage assessment, and platform service life prediction.

[0019] Offshore wind farms operate in complex and variable environments. Extreme wind and wave events (such as sudden strong winds, giant waves, and typhoon aftershocks) often exert a strong impact on wind turbine towers, foundations, submarine cables, and maintenance platforms within a short period. Real-time prediction and early warning of extreme sea conditions are crucial technical aspects for ensuring the safety of wind power equipment and reducing unplanned downtime. Traditional numerical models, such as the Shallow Wave Model (SWAN) and Weather Research and Forecasting (WRF) models, while physically rigorous, rely on external background fields and struggle to quickly respond to locally abrupt wind and wave dynamic conditions. They also suffer from long computation times and difficulty in real-time updates. Statistical regression models and neural network regression models (such as LSTM and GRU), while capable of handling some time-series features, tend to underestimate extreme values ​​when extreme samples are scarce and wind and wave nonlinear coupling is significant, leading to warning failures or delays.

[0020] To address the aforementioned technical problems, this invention proposes a sea state prediction method. The implementation details of the self-labeled semi-supervised training method for the classification model in this embodiment are described below. The following content is only for ease of understanding and is not necessary for implementing this solution.

[0021] Example 1: like Figure 1 As shown, this application provides a method for predicting sea states. This method is applicable to electronic production equipment, which can be a server, mobile terminal, computer, cloud platform, etc. The data processing functionality of the production equipment provided in this application embodiment can be implemented by the processor of the electronic production equipment calling program code, wherein the program code can be stored in a computer storage medium. The sea state prediction method includes: Step S1: Obtain the wave characteristics and wind field characteristics of the target sea area within a preset time range.

[0022] In some embodiments, step S1, "acquiring wave characteristics and wind field characteristics of the target sea area within a preset time range," includes: Step S11: Use microwave radar to acquire the sea surface echo spectrum of the target sea area within a preset time range.

[0023] Microwave radar is a non-contact sea surface spectral inversion device based on the principle of electromagnetic scattering, typically operating in the 24–26 GHz range. It obtains the spectral distribution of the reflected sea surface signal by transmitting continuous wave or frequency-modulated continuous wave signals and receiving the echoes formed by Brillouin scattering from the sea surface. At the signal processing level, wave microwave radar can calculate the main wave elements in real time through Doppler frequency shift analysis and spectral inversion algorithms. Youyi Bogao ( ): A statistical measure representing the total energy of ocean waves; Main wave period ( ): Reflects the dominant period of the wave energy concentration region; Wave direction ( : Indicates the direction of propagation of the main wave system.

[0024] Because it uses a non-contact measurement method, it is unaffected by wave surface drift or sensor immersion, and can maintain high-precision continuous observation even in extreme sea conditions. Microwave wave radar is usually installed on the deck of a booster station or the deck of a mother ship, enabling continuous monitoring of the local sea surface of wind farms.

[0025] In the time domain, wave radar outputs high-frequency (typically at 1–5 minute intervals) continuous spectral time-series data, which not only reflects mean sea state characteristics but also reveals wave group evolution, spectral energy migration, and energy amplification phenomena caused by wind and wave superposition. This high-temporal-resolution spectral sequence information is particularly crucial for extreme event prediction, as it can capture low-frequency energy accumulation or spectral peak multimodal characteristics tens of minutes before extreme wave formation, providing criteria for early warning.

[0026] In addition, wave radar data contains rich high-order statistical features, such as spectral peak migration rate, energy concentration, and directional diffusion angle, which can be further extracted through time-frequency analysis methods (such as short-time Fourier transform (STFT) and wavelet transform (CWT)) to provide more physical feature representations for subsequent deep learning model inputs.

[0027] Step S12: Determine the wave characteristics of the target sea area at various times within a preset time range based on the sea surface echo map.

[0028] The wave characteristics in this application include significant wave height, dominant wave period, and dominant wave direction, wherein the raw data provided by the K-band wave radar is a sea surface echo spectrum. The main wave elements can be obtained by inverting the spectrum energy integral.

[0029] (1) Meaningful wave high ( ) (1) Spectrum function (unit) (), obtained by integrating the directional spectrum; Wave frequency ( ); The zeroth moment of the spectrum represents the total wave energy.

[0030] (2) Main wave period ( ) (2) Spectral energy density The peak frequency corresponds to the main energy frequency.

[0031] (3) Direction of the main wave ( ) (3) : Directional spectrum function (the joint distribution of wave energy with respect to frequency and direction); Wave direction angle (°); : The direction of propagation of the dominant wave.

[0032] Furthermore, by using Equation 1-3, the wave characteristics at each moment within the preset time range can be calculated based on the sea surface echo spectrum collected by the microwave radar at each moment within the preset time range.

[0033] Step S13: Use laser wind radar to obtain the wind speed field at multiple radial points in front of the wind turbine impeller within a preset time range.

[0034] A laser wind radar is a wind field remote sensing device based on the Doppler laser scattering principle. Its working principle is as follows: the laser radar emits laser beams at different azimuth and elevation angles. The beams are scattered by aerosol particles in the atmosphere, and the frequency shift of the echo signal is proportional to the wind speed component along the beam direction. By inverting multiple echoes, a spatially distributed three-dimensional wind speed vector field can be obtained.

[0035] In offshore wind power applications, nacelle-mounted wind measuring radars are typically installed on top of the wind turbine nacelle, facing directly in front of the rotor, with a typical detection range of 300–400 m. Compared to traditional wind tower or blade-mounted sensors, this radar can directly measure the inflow wind field in front of the turbine rotor, offering advantages such as non-contact operation, high safety, and strong real-time performance. Its observation data includes the following main physical quantities: Horizontal wind speed and direction sequence ; Turbulence intensity ( ): Describes the intensity of airflow fluctuations; Turbulent dissipation rate ( ): Reveals the relationship between energy cascades and the intensity of turbulent activity; Horizontal wind shear and vertical wind shear: reflect the stratification characteristics of the wind field and are important driving forces for wind-wave coupling.

[0036] These wind field parameters can reveal the non-uniformity and transient characteristics of the atmospheric boundary layer structure. Extensive field measurements show that when wind shear increases or turbulent dissipation rate rises, the spectral energy density and directional stability of sea surface waves change significantly, often triggering extreme waves or abrupt changes in wave direction. Therefore, the high-frequency wind field sequences provided by nacelle-mounted laser wind radar are crucial observational information for characterizing the formation mechanisms of extreme sea states.

[0037] Step S14: Determine the wind field characteristics of the target sea area at various times within a preset time range based on the wind speed field.

[0038] The wind field characteristics in this application include: average horizontal wind speed and direction, turbulence intensity, horizontal and vertical wind shear, and turbulence dissipation rate. The nacelle-type laser anemometer can acquire the wind speed field at multiple radial measurement points in front of the wind turbine impeller, and the following wind field characteristics are obtained from the wind speed field inversion: (1) Average wind speed and wind direction (4) : No. Wind speed at each sampling point; Corresponding wind direction; : Sample size.

[0039] (2) Turbulence intensity ( ) (5) Standard deviation of wind speed ( ); Average wind speed ( ).

[0040] (3) Horizontal and vertical wind shear (6) Wind speed ( ); Horizontal distance ( ); :high( ); : These represent the wind speed gradients in the horizontal and vertical directions, respectively.

[0041] (4) Turbulent dissipation rate ( ) (7) : Empirical constant (taken as 0.7–0.9); Standard deviation of turbulent velocity; : Turbulent integral scale ( ); : Turbulent energy dissipation rate ( This reflects the trend of turbulence intensity variation.

[0042] Furthermore, Equation 4-7 can be used to calculate the wind field characteristics at each moment within the preset time range based on the wind speed field collected by the laser wind measuring radar at each moment within the preset time range.

[0043] Step S2: Use a bidirectional gated loop unit to extract wave time series features and wind field time series features from the wave features and wind field features within a preset time range, respectively.

[0044] In some embodiments, step S2, "extracting wave time-series features and wind field time-series features from the wave features and wind field features within a preset time range using a bidirectional gated loop unit," includes: Step S21: Time-align the wave features and the wind field features.

[0045] Due to the difference in sampling frequencies between the two types of radar (wave radar typically uses 1–2), Sampling, LiDAR is 1–10 First, time unification and interpolation correction are required. Linear interpolation is used to resample all data to a uniform time step. (This article is set to 1 minute): (8) Variable description: : A unified identifier for the interpolation results of wind and wave characteristic variables at time t. Here, wind and wave characteristics refer to ocean wave characteristics or wind field characteristics. , : Valid data points of the original feature variable observation data at adjacent time points; , : The corresponding timestamp; Target sampling interval (1 minute).

[0046] Equation 8 above illustrates the interpolation process for ocean wave features; the same applies to wind field features. After interpolating the wind field and ocean wave features using Equation 8, time alignment of the wind field and ocean wave features is achieved. However, in practice, although the wind field and ocean wave features have been time-aligned and can proceed to the next step of time series modeling, considering that these features are directly derived from the collected data, there will be interfering data. Therefore, data cleaning of the wind field and ocean wave features is necessary. The cleaning process includes: missing data and abrupt value correction, normalization, and anomaly detection. Equations 9 and 10 below demonstrate missing data and abrupt value correction and normalization using ocean wave features as an example. Similarly, Equations 9 and 10 can also be used for missing data and abrupt value correction and normalization of wind field features, requiring only simple parameter substitution.

[0047] Missing values ​​and mutation value correction If the missing data period is less than a threshold (e.g., 10 minutes), linear interpolation is used for patching; if the continuous missing data period exceeds the threshold, data for that time period are simultaneously removed from both wind field and wave features. Abrupt changes are detected using a sliding window constraint based on the mean and standard deviation. (9) It is considered an anomaly and replaced with the neighborhood mean.

[0048] Window length is The moving average; : Corresponding moving standard deviation; : Sliding window length (30 minutes in this article, i.e.) ).

[0049] Normalization All features were normalized to the [0,1] interval to improve the stability of neural network training. To eliminate the influence of different units on network training, all features were subjected to min-max normalization: (10) : Dimensionless data after normalization; , : The minimum and maximum values ​​of this feature in the training set.

[0050] The anomaly detection in this application includes two steps: statistical anomaly detection and wind and wave physical consistency constraints.

[0051] Because wave radar signals are affected by sea surface scattering angle, raindrop interference, and equipment calibration errors, they are prone to non-physical abrupt changes in wave height or direction; wind radar may produce false wind speed inversions due to excessively low signal strength. To eliminate data that does not conform to physical laws or is caused by instrument malfunctions, both statistical and physical constraints are introduced.

[0052] (1) Z-score statistical anomaly detection (11) when When the value is (preset value), it is judged as an outlier and removed.

[0053] : A single sample value of the wind and wave characteristic variable, where wind and wave characteristics refer to ocean wave characteristics or wind field characteristics; : Feature mean; Characteristic standard deviation.

[0054] (2) To avoid the deep learning model learning non-physical associations, physical consistency detection is performed on the repaired data.

[0055] Wind and wave physical consistency constraints (12) : Meaningful wave height (m); Wind speed ( ); Gravitational acceleration (9.81) ); Empirical coefficient (0.016–0.02).

[0056] This constraint is used to determine whether the wave inversion results are consistent with the wind energy input. If the results exceed the upper limit of the calculation results on the right, they are marked as abnormal wave heights.

[0057] The above process completes the data processing of wave and wind field characteristics, resulting in reliable and time-aligned wave and wind field characteristics.

[0058] Step S22: Input the time-aligned wave features and the wind field features into the bidirectional gated loop unit respectively.

[0059] Step S23: Obtain the wave timing characteristics and the wind field timing characteristics through the bidirectional gated loop unit.

[0060] Before being input into the bidirectional gated loop unit, the wave characteristics and wind field characteristics need to be processed to meet the input conditions of the bidirectional gated loop unit.

[0061] The input to the bidirectional gated recurrent unit is the multi-source feature sequence from the previous n steps. .

[0062] Will (Wave characteristics) Input into the wave channel and... (Wind field characteristics) Input into the wind field channel: • Include and spectral characteristics; • Include .

[0063] Before inputting into the bidirectional gated recurrent unit, the wave features and wind field features need to be processed by a 1D convolution (or linear mapping) + layer normalization (LayerNorm) to obtain a fixed-dimensional representation.

[0064] • Recommendation: Kernel size The output dimension is d=128.

[0065] When obtaining the temporal characteristics of ocean waves and wind fields through a bidirectional gated cyclic unit (Bi-GRU), the update formula of the Bi-GRU is as follows: (13) Variable definition: : A single sample value of the wind and wave characteristic variable, where wind and wave characteristics refer to ocean wave characteristics or wind field characteristics; The current hidden state; , Reset the door and update the door; Weight matrix; : Bias term; : Sigmoid function; Hadamard element-wise multiplication.

[0066] Bi-GRU has fewer parameters than LSTM, resulting in higher training efficiency. Furthermore, its bidirectional structure enables simultaneous modeling of historical and future trends, enhancing the model's sensitivity in identifying precursors to extreme events.

[0067] Recommendation: Bidirectional GRU (2 layers, hidden=128).

[0068] Step S3: Determine the spatiotemporal attention features of the ocean waves and the spatiotemporal attention features of the wind field based on the ocean wave temporal features and the wind field temporal features, respectively.

[0069] In some embodiments, step S3, "determining the spatiotemporal attention features of the ocean waves and the spatiotemporal attention features of the wind field based on the ocean wave time series features and the wind field time series features respectively," includes: Step S31: Calculate the wave attention weights of the wave temporal features at each time point.

[0070] Step S32: Determine the wave spatiotemporal attention features based on the wave temporal characteristics and wave attention weights at each time point.

[0071] Step S33: Calculate the wind field attention weights of the wind field temporal features at each time point.

[0072] Step S34: Determine the spatiotemporal attention features of the wind field based on the temporal features of the wind field at each time point and the attention weight of the wind field.

[0073] During the wind-wave coupling process, the importance of different time steps and feature dimensions changes dynamically. To enhance the model's attention to critical moments (such as sudden increases in wind speed or wave height), this paper introduces a spatiotemporal attention mechanism.

[0074] The formula for calculating attention weights is: (14) Variable definition: Time step Attention score; Normalized attention weights; Trainable parameters; : The hidden state at the current moment in Equation 13; : A single sample value of the wind and wave characteristic variable, where wind and wave characteristics refer to ocean wave characteristics or wind field characteristics.

[0075] The weighted output is: (15) This means focusing attention weights on key time periods to improve the model's responsiveness during sudden storms.

[0076] Note that this step involves calculating the attention weight within each channel (wave channel and wind channel), and aggregating the time-important time periods. , Represents the spatiotemporal attention characteristics of ocean waves. This represents the spatiotemporal attention characteristics of the wind field.

[0077] Step S4: Determine the wind-wave fusion features based on the spatiotemporal attention features of the ocean waves and the spatiotemporal attention features of the wind field.

[0078] There is a significant physical coupling relationship between wind and waves: strong winds increase wave height, and waves, in turn, alter the wind field structure. To characterize this nonlinear interaction, this paper designs an interactive fusion layer that achieves wind and wave data fusion through feature cross-mapping.

[0079] (16) Variable definition: Characteristics of wind and wave fusion; Spatiotemporal attention characteristics of ocean waves; Spatiotemporal attention characteristics of wind fields; Vector concatenation operation; Weights and biases of fully connected layers; : Non-linear activation function (ReLU).

[0080] Features after fusion It contains both wind and wave coupling information and key time series patterns, and serves as the input to the final prediction layer.

[0081] Step S5: Extract extreme sensing features from the wind and wave fusion features using the extreme sensing module.

[0082] In some embodiments, step S5, "extracting extreme sensing features from the wind and wave fusion features using the extreme sensing module," includes: Step S51: Calculate the extreme probability score of the wind and wave fusion feature using a multilayer perceptron network.

[0083] Step S52: Determine the extreme weights based on the extreme probability scores.

[0084] Step S53: Use a nonlinear amplification function to amplify the saliency features of the wind and wave fusion feature to obtain the fusion amplification feature.

[0085] Step S54: Determine the extreme sensing feature based on the fusion amplification feature, the wind and wave fusion feature, and the extreme weight.

[0086] In extreme sea state prediction, traditional deep learning models (such as LSTM, GRU, or Transformer) typically suffer from two significant problems: Imbalanced sample size: The number of extreme sea state event samples is far less than that of normal samples. The model is easily dominated by the "normal state" during training, resulting in a weak ability to identify extreme samples.

[0087] Insufficient characteristic energy: The dynamic response of wind and waves under extreme conditions (such as sudden rise in wave height and rapid change in wind shear) has nonlinear enhancement characteristics. Without additional significant amplification, the model is unable to capture its abrupt change signal.

[0088] Therefore, this paper introduces an Extreme Awareness Module (EAM) after the cross-modal feature fusion layer. Its goal is to improve the model's sensitivity to extreme events and prediction accuracy by adaptively weighting and amplifying extreme samples, based on the joint wave-wind field features. EAM operates on the fused wave and wind field feature vector. .

[0089] The overall process is as follows: (17) in, Represents the extreme sensing mapping function; For extreme sample weight vectors; The characteristic magnification factor, It is a regular balance term.

[0090] (1) Calculation of extreme sample weights The model first calculates the extreme probability score based on the wind and wave conditions of the current sample. This is achieved through a lightweight multilayer perceptron (MLP) network: (18) in: The Sigmoid function maps the output to the interval [0,1]. Extreme probability mapping matrix; : Bias term; : The predicted probability that the current sample is in extreme sea conditions.

[0091] Subsequently, extreme weights are defined based on this probability. : (19) in: Extreme sample augmentation coefficient (typically ranging from 0.5 to 2.0), which controls the model's sensitivity to amplification of extreme samples.

[0092] when When it is close to 0 (normal state). ; when When it approaches 1 (extreme state). That is, amplifying the characteristic response.

[0093] (2) Significant amplification To enhance the model's response to abrupt signals (such as sudden increases in wave height or abrupt changes in wind speed), EAM introduces a learnable nonlinear amplification function: (20) in: : Input wind and wave fusion features; : Amplified layer weight matrix; Hyperbolic tangent activation function, used to suppress noise while preserving extreme value characteristics; Feature magnification factor (usually between 0.1 and 0.5), controls the intensity of feature magnification; Element-wise multiplication operation; Significant features after magnification.

[0094] This operation is equivalent to weighting and strengthening the high-amplitude mutation part of the feature vector, making the features of extreme samples more identifiable in subsequent classification or regression stages.

[0095] (3) Dynamic reweighted output The final output of the extreme perceptual features is: (twenty one) in: Regularization balancing factor, which controls the fusion ratio of enhanced features and original features; Extreme perception features are used in subsequent regression and classification heads.

[0096] when When the size is large, EAM tends to use augmentation features to improve its ability to identify extreme cases; when When the size is small, the model focuses more on overall stability and physical consistency.

[0097] Step S6: Predict the sea conditions of the target sea area based on the extreme sensing characteristics. In some embodiments, step S6, "predicting the sea state of the target sea area based on the extreme sensing characteristics," includes: Step S61: Use a two-layer fully connected network to predict the wave parameters of the target sea area in future time periods based on the extreme sensing features.

[0098] Step S62: Predict the probability of extreme events based on the extreme perception features using a single-layer fully connected network and an activation function.

[0099] Step S63: Obtain the preset probability threshold.

[0100] Step S64: Determine whether to issue an extreme event warning based on the extreme event probability and the probability threshold.

[0101] This paper adopts a multi-task joint learning architecture, setting regression prediction heads and classification prediction heads at the same network end, both sharing the extreme perceptual features output by EAM. Meanwhile, the shared weights are optimized to improve generalization ability.

[0102] (1) The prediction layer structure is implemented using a two-layer fully connected network: (twenty two) in: : Weight matrix of fully connected layer; : Bias term; The function enhances the ability to extract nonlinear features.

[0103] : Future sea state prediction vector.

[0104] The classification prediction head uses a single-layer fully connected layer with a Sigmoid function: (twenty three) in: : The predicted probability of extreme events; The Sigmoid function maps the output to [0,1]. Extreme probability mapping matrix; like If so, it is determined that extreme sea conditions may occur in the future. This is a preset probability threshold.

[0105] (2) When training the prediction layer, a joint loss function is used to simultaneously optimize regression and classification performance: (twenty four) in: Weighting coefficients (usually 1:1 or 2:1); Prediction error for continuous variables; : Classification error under extreme conditions.

[0106] The regression loss uses weighted mean squared error (MSE): (25) in: The weights are dynamically adjusted based on the extreme probabilities output by EAM to ensure that extreme samples are reinforced during training. Number of training samples; : The continuous output vector predicted by the model; : The vector of actual measured or labeled values; : L2 norm squared (Euclidean distance squared); Extreme sample magnification factor; : Sample weighting coefficient; : The probability that the i-th sample in EAM is classified as an extreme state.

[0107] The classification loss uses binary cross-entropy (BCE): (26) in, : A label indicating whether the sample represents an extreme state; : The extreme probabilities predicted by the model.

[0108] The warning trigger logic occurs during the model prediction phase, and the system outputs: (27) An alert is triggered when the following conditions are met: (28) in: : Sea state indicators for future time steps; The height of the wave in the future time step; The wind speed in the future time step; : Turbulence intensity at future time steps; : Predicted time (30 minutes); Extreme probability threshold (taken as 0.7 here); Wave height threshold (taken as 5.0 m here); Wind speed threshold (taken as 25 m / s here); : Turbulence intensity threshold.

[0109] The method in this application can be implemented using a deep learning model that fuses multiple models, such as... Figure 2 As shown, the deep learning model formed by the fusion in this application consists of six parts: a feature embedding layer, a temporal coding layer, a spatiotemporal self-attention module, an interactive fusion layer, an extreme-aware module, and a multi-task joint learning prediction layer. The model adopts a dual-channel structure to extract the temporal dynamics and spatial correlation of wave and wind field features respectively. Then, a self-attention mechanism is introduced into each channel to enhance the model's perception of abnormal state changes. Next, an interactive fusion layer (CFL) is introduced to achieve nonlinear coupling mapping between the wave and wind field features output from the two channels, forming a wind-wave fusion feature. The introduced extreme-aware module (EAM) can improve the model's sensitivity to extreme events and prediction accuracy through adaptive weighting and energy amplification mechanisms for extreme samples. Finally, a multi-task learning (MTL) architecture is adopted, with a regression head and a classification head set at the same network end to achieve early prediction of short-term sea state indicators (such as meaningful wave height, wind speed, etc.).

[0110] This invention proposes a sea state prediction method to address the problems of inaccurate prediction and delayed response of extreme wind and wave events in existing offshore wind farms. It utilizes real-time monitoring data from microwave wave radar and laser wind radar to construct a high-dimensional time-series feature system coupled with wind and waves, enabling early identification and probabilistic forecasting of extreme sea states. Through data quality control and feature construction, feature parameters such as sea surface wave height, wave direction, wave period, and wind speed, wind direction, turbulence intensity, and wind shear are extracted to form a unified spatiotemporal feature input set. This invention employs a deep learning model integrating convolutional neural networks, gated recurrent units, and Bi-GRU structures, and introduces an Extrema Awareness Module (EAM) to achieve dynamic weighted learning of rare extreme samples. The model outputs continuous physical quantity predictions and extreme event probabilities in a multi-task manner, achieving integrated sea state trend prediction and intelligent early warning of extreme conditions. Compared with traditional prediction methods based on statistics or single-source measurements, this method significantly improves the accuracy, timeliness, and deployability of extreme sea state identification, making it suitable for real-time operation monitoring and risk control systems for offshore wind farms.

[0111] Example 2: Based on the foregoing embodiments, this application provides a sea state prediction device. The various modules and units included in the device can be implemented by a processor in a computer device; of course, they can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0112] like Figure 3 As shown, a sea state prediction device includes: a first acquisition module 1, a first execution module 2, a first determination module 3, a second determination module 4, a second execution module 5, and a first prediction module 6.

[0113] The first acquisition module 1 is used to acquire wave characteristics and wind field characteristics of the target sea area within a preset time range. The first execution module 2 is used to extract wave temporal characteristics and wind field temporal characteristics from the wave characteristics and wind field characteristics within the preset time range using a bidirectional gated loop unit. The first determination module 3 is used to determine wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics based on the wave temporal characteristics and wind field temporal characteristics, respectively. The second determination module 4 is used to determine wind-wave fusion characteristics based on the wave spatiotemporal attention characteristics and wind field spatiotemporal attention characteristics. The second execution module 5 is used to extract extreme perception characteristics from the wind-wave fusion characteristics using an extreme perception module. The first prediction module 6 is used to predict the sea conditions of the target sea area based on the extreme perception characteristics.

[0114] The various modules in the aforementioned sea state prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the device in hardware form or independently of it, or stored in the memory of the processing device in software form, so that the processor can call and execute the operations corresponding to each module. It should be noted that the module division in this embodiment is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods.

[0115] Example 3: Thirdly, this application provides a computer electronic production device, such as... Figure 4 As shown, it includes: at least one processor 901; and a memory 902 communicatively connected to the at least one processor 901; wherein the memory 902 stores instructions executable by the at least one processor 901, the instructions being executed by the at least one processor 901 to enable the at least one processor 901 to execute a drilling early warning method in the above embodiments.

[0116] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0117] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0118] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0119] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0120] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0121] The processor can communicate with external devices via the I / O bus through wired or wireless networks.

[0122] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0123] Example 4: Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0124] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Computer-readable storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, and computer storage media (e.g., hard disks, floppy disks, solid-state drives, removable disks, CD-ROMs, DVD-ROMs, Blu-ray discs, etc.).

[0125] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0126] Example 5: Fifthly, this application proposes a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method described in any of the first aspects.

[0127] The data query system in this application is capable of communicating with various existing systems. Moreover, the data query system has preset data processing rules for the data in each system, which are used to convert the raw data queried from each system into data in a unified format and display it to the user. The system in this application has multiple communication protocols, which can meet the needs of communicating with multiple systems.

[0128] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0130] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0131] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0132] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0133] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a controller to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0134] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein.

[0135] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for predicting sea states, characterized in that, include: Acquire the wave and wind field characteristics of the target sea area within a preset time range; The temporal features of ocean waves and wind fields are extracted from the ocean wave features and wind field features within a preset time range using a bidirectional gated loop unit. The spatiotemporal attention features of ocean waves and the spatiotemporal attention features of wind fields are determined based on the ocean wave temporal features and the wind field temporal features, respectively. The wind-wave fusion characteristics are determined based on the spatiotemporal attention characteristics of the ocean waves and the spatiotemporal attention characteristics of the wind field. Extreme sensing features are extracted from the wind and wave fusion features using an extreme sensing module; The sea conditions of the target sea area are predicted based on the extreme sensing characteristics.

2. The method according to claim 1, characterized in that, The acquisition of wave and wind field characteristics of the target sea area includes: The sea surface echo spectrum of the target sea area within a preset time range is obtained using microwave radar; The wave characteristics of the target sea area at various times within a preset time range are determined based on the sea surface echo spectrum. The wind speed field at multiple radial points in front of the wind turbine impeller within a preset time range is obtained using a laser wind radar. The wind field characteristics of the target sea area at various times within a preset time range are determined based on the wind speed field.

3. The method according to claim 1, characterized in that, The extraction of ocean wave temporal features and wind field temporal features from the ocean wave features and wind field features within a preset time range using a bidirectional gated loop unit includes: The wave characteristics and wind field characteristics are time-aligned; The time-aligned wave characteristics and the wind field characteristics are respectively input into the bidirectional gated loop unit; The wave timing characteristics and the wind field timing characteristics are obtained through the bidirectional gated loop unit.

4. The method according to claim 1, characterized in that, The determination of the spatiotemporal attention features of ocean waves and the spatiotemporal attention features of wind fields based on the ocean wave time series features and the wind field time series features respectively includes: Calculate the wave attention weights for the wave temporal features at each time point; The spatiotemporal attention features of the waves are determined based on the temporal characteristics of the waves at each time point and the attention weights of the waves. Calculate the wind field attention weights for the wind field temporal features at each time point; The spatiotemporal attention features of the wind field are determined based on the temporal characteristics of the wind field at each time point and the attention weight of the wind field.

5. The method according to claim 1, characterized in that, The extraction of extreme sensing features from the wind and wave fusion features using the extreme sensing module includes: The extreme probability scores of the wind and wave fusion features are calculated using a multilayer perceptron network; The extreme weights are determined based on the extreme probability scores. The saliency of the wind-wave fusion feature is amplified using a nonlinear amplification function to obtain the fusion amplification feature; The extreme perception feature is determined based on the fusion amplification feature, the wind and wave fusion feature, and the extreme weight.

6. The method according to claim 1, characterized in that, The prediction of sea conditions in the target sea area based on the extreme sensing characteristics includes: A two-layer fully connected network is used to predict the wave parameters of the target sea area in future time periods based on the extreme sensing characteristics; The probability of extreme events is predicted based on the extreme perception features using a single-layer fully connected network and an activation function. Obtain the preset probability threshold; Whether to issue an extreme event warning is determined based on the extreme event probability and the probability threshold.

7. A sea state prediction device, characterized in that, include: The first acquisition module is used to acquire the wave characteristics and wind field characteristics of the target sea area within a preset time range; The first execution module is used to extract wave time-series features and wind field time-series features from the wave features and wind field features within a preset time range using a bidirectional gated loop unit. The first determining module is used to determine the spatiotemporal attention features of the ocean waves and the spatiotemporal attention features of the wind field based on the ocean wave temporal features and the wind field temporal features, respectively. The second determining module is used to determine the wind-wave fusion features based on the spatiotemporal attention features of the ocean waves and the spatiotemporal attention features of the wind field. The second execution module is used to extract extreme perception features from the wind and wave fusion features using the extreme perception module; The first prediction module is used to predict the sea conditions of the target sea area based on the extreme perception features.

8. A computer electronic production equipment, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images