Short-term sea wave significant wave height prediction method based on MGSNet
By screening, decomposing, and constructing the MGSNet model, the problem of insufficient accuracy of machine learning models in short-term effective wave height prediction is solved. It achieves effective characterization of the relationship and time dependence of input variables, improves prediction accuracy, and is applicable to offshore platform operations and shipping risk management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA THREE GORGES UNIV
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-19
AI Technical Summary
Existing machine learning models have low accuracy in predicting short-term effective wave heights, struggle to effectively characterize the complex relationships between multiple input variables and their significant time dependencies, and are insufficiently adapted to local abrupt changes.
The correlation coefficient method was used to screen marine environmental data, multivariate empirical mode decomposition was performed, an MGSNet model was constructed for training and prediction, and inverse normalization reconstruction was performed to improve prediction accuracy.
By characterizing the complex relationships and time dependencies between input variables, the prediction accuracy of effective wave height on the hourly scale is improved, making it easier to apply in offshore platform operations and shipping risk management.
Smart Images
Figure CN122064951A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of significant wave height prediction technology, and specifically to a short-term significant wave height prediction method based on MGSNet. Background Technology
[0002] Significant wave height (SWH) is a key parameter characterizing wave energy and sea state risk levels, and its short-term forecast accuracy has a significant impact on the safety of offshore platform operations and the efficiency of maritime cargo transportation. For offshore platform operations, the operational window is typically highly sensitive to sea state thresholds, especially to hourly fluctuations in SWH. Failure to identify rising trends or sudden changes in SWH in advance on an hourly scale can easily lead to unreasonable operational planning and resource scheduling, and frequent operational interruptions. For maritime cargo transportation, insufficient short-term forecast accuracy weakens the effectiveness of dynamic optimization of vessel routes, energy consumption control, and safety and hazard avoidance decisions, thereby increasing navigation risks and reducing transportation efficiency.
[0003] Traditional methods for predicting short-term significant wave height (SSW) mainly include empirical statistical methods and numerical simulation methods. Empirical statistical methods typically build models based on simplifying assumptions and historical statistical patterns, offering high computational efficiency, but their prediction performance is poor in complex nonlinear sea states or sudden events. Numerical simulation methods establish governing equations based on wave physics mechanisms, providing relatively reliable prediction results, but they are highly dependent on parameter settings and computational resources, making real-time deployment difficult in some scenarios. Therefore, traditional methods still struggle to balance accuracy and efficiency in SSW prediction. Machine learning techniques have been introduced to improve short-term prediction performance. These methods do not require theoretical assumptions and can learn the nonlinear mapping between input and output through sample data, thereby improving prediction accuracy. However, existing machine learning models still face challenges: on the one hand, ocean parameters exhibit complex relationships and significant time dependencies, making it difficult for models to fully learn the relationships between different input variables; on the other hand, short-term sequences have non-stationarity and multi-scale fluctuation characteristics, making models insufficiently adaptable to local abrupt changes, thus limiting the prediction accuracy of SSW. Summary of the Invention
[0004] The technical problem of this invention is to solve the problems of low accuracy in short-term effective wave height prediction of existing machine learning models, difficulty in effectively representing the complex relationship between multiple input variables and their significant time dependence.
[0005] The purpose of this invention is to solve the above-mentioned problems and propose a short-term wave significant height prediction method based on MGSNet, comprising the following steps:
[0006] S1: Acquire effective wave height data and marine environmental data using buoys, and perform preprocessing; S2: The correlation coefficient method is used to calculate the correlation coefficient between marine environmental data and significant wave height, and the time series dataset is obtained after screening and preprocessing. S3: Perform multivariate empirical mode decomposition (MEMD) on the time series dataset to obtain intrinsic mode functions (IMF) and residuals; S4: Construct the MGSNet model, preprocess the intrinsic mode functions (IMF) and residuals obtained from the decomposition, and input them into the model for training and prediction to obtain the prediction results; S5: Perform inverse normalization and reconstruction on the prediction results to obtain the effective wave height prediction value.
[0007] Preferably, in step S1, the preprocessing includes performing time resolution unification processing on the acquired valid wave height data and marine environmental data, and performing outlier removal and time alignment processing on the data.
[0008] Furthermore, in step S2, the correlation coefficient method is used to calculate the correlation coefficient between marine environmental data and significant wave height. The calculation formula is as follows: ; In the formula, E( ) represents the expected value, S represents the time series of the significant wave height, and O represents the time series of any input variable.
[0009] Preferably, step S3 includes the following sub-steps: 1) Perform Multivariate Empirical Mode Decomposition (MEMD) on the time series dataset to obtain a multivariate time series. The expression for the multivariate time series is: ; In the formula, Indicates the first A marine environmental variable at time Time series data, Indicates the number of variables; 2) Decompose the multivariate time series X(t) into several intrinsic mode functions and residuals, expressed as: ; In the formula, Indicates the first Eigenmode function vector, This indicates the number of IMFs obtained from the decomposition. Represents the residual.
[0010] Furthermore, for the intrinsic mode function, the difference between the number of local extrema and the number of zero crossings of each component does not exceed 1; the local mean of the upper and lower envelopes of each component is zero or approximately zero over the entire sequence; and each component satisfies the condition that its local mean is zero.
[0011] Preferably, in step S4, the MGSNet model includes a gated loop unit, which uses two gate structures to update and transmit information; the update gate is responsible for controlling the influence of the previous state information on the current state, and the reset gate is responsible for controlling the ignoring of the previous state information.
[0012] Preferably, in step S5, the prediction result is inversely normalized, and the calculation formula is: ; In the formula, and They represent the first The maximum and minimum values of each decomposed component data Indicates the first Normalized predicted values of each decomposed component, Indicates the i-th after denormalization Predicted values for each decomposed component.
[0013] In step S5, the reconstruction involves reconstructing the predicted effective wave height in the normalized space, calculated as follows: ; In the formula, and These represent the maximum and minimum values in the effective wave height time series data before normalization, respectively. This represents the sum of the predicted values of each decomposition component after inverse normalization.
[0014] Compared with the prior art, the beneficial effects of the present invention include: 1) The short-term wave significant height prediction method based on MGSNet proposed in this invention improves the prediction accuracy of hourly wave significant height by screening the correlation of multivariate marine environmental sequences and obtaining components at different scales through multivariate empirical mode decomposition, and then using the MGSNet model for prediction and reconstruction.
[0015] 2) The MGSNet model proposed in this invention can characterize the complex relationships and significant time dependencies between input variables under multivariate input conditions, thereby more accurately depicting the short-term variation characteristics of the effective wave height.
[0016] 3) This invention constructs a time series prediction process based on buoy observation data. The prediction output is reconstructed and denormalized to obtain the final effective wave height result, which is convenient for application in scenarios such as safety assurance of offshore platform operations and shipping risk management. Attached Figure Description
[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0018] Figure 1This is a schematic diagram of the short-term significant wave height prediction method based on MGSNet according to an embodiment of the present invention. Figure 2 This is a schematic diagram showing the correlation coefficient between marine environmental data and significant wave height in an embodiment of the present invention. Figure 3 This is a schematic diagram of the results of the first 10 intrinsic mode functions of the multivariable empirical mode decomposition in an embodiment of the present invention; Figure 4 This is a schematic diagram of the residual intrinsic modes and residual results of multivariable empirical mode decomposition according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the gated loop unit structure according to an embodiment of the present invention; Figure 6 This is a schematic diagram comparing the time series of effective wave height prediction for the next 1 hour according to an embodiment of the present invention. Figure 7 This is a schematic diagram comparing the local time sequence of the effective wave height prediction for the next 1 hour according to an embodiment of the present invention. Detailed Implementation
[0019] like Figure 1 As shown, a short-term wave significant height prediction method based on MGSNet is proposed. S1: Obtain significant wave height data and marine environmental data using buoys and perform preprocessing.
[0020] Preprocessing includes unifying the time resolution of the acquired valid wave height data and marine environmental data, and removing outliers and aligning the data over time.
[0021] Specifically, buoys with complete data storage were selected as research objects, and the corresponding significant wave height data and other marine environmental data were extracted. The original 0.5h time resolution was converted to 1h time resolution.
[0022] S2: The correlation coefficient method is used to calculate the correlation coefficient between marine environmental data and significant wave height, and the time series dataset is obtained after screening and preprocessing.
[0023] The correlation coefficient method is used to calculate the correlation coefficient between marine environmental data and significant wave height. The formula is as follows: ; In the formula, E( ) represents the expected value, S represents the time series of the significant wave height, and O represents the time series of any input variable.
[0024] like Figure 2 As shown, the filtered time series dataset contains four variables: significant wave height, wind speed, peak gust wind speed, and average wave period.
[0025] S3: Perform multivariate empirical mode decomposition (MEMD) on the time series dataset to obtain intrinsic mode functions (IMF) and residuals.
[0026] The filtered time series dataset is normalized using the following formula: ; In the formula, Representing time series data, This represents the normalized time series data. and These represent the maximum and minimum values of the time series data, respectively.
[0027] Multivariate Empirical Mode Decomposition (MEMD) is performed on the time series dataset to obtain a multivariate time series. The expression for the multivariate time series is: ; In the formula, Indicates the first A marine environmental variable at time Time series data, Indicates the number of variables; Multivariate time series Decomposed into several intrinsic mode functions and residuals, the expression is: ; In the formula, Indicates the first Eigenmode function vector, This indicates the number of IMFs obtained from the decomposition. Represents the residual.
[0028] Define decomposition variables Multivariate time series The formula for calculation is: ; ; In the formula, Indicates the first Time series of each decomposed component Indicates the index of the decomposed variable. This represents the total number of intrinsic mode functions (IMFs) obtained from the decomposition. This represents the stationary vector remaining after the original signal is decomposed. Represents a time variable.
[0029] The intrinsic mode function, where the difference between the number of local extrema and the number of zero crossings in each component does not exceed 1, is expressed as: ; In the formula, This represents the total number of local maxima and local minima. It represents the number of points that cross zero.
[0030] The local mean of the upper and lower envelopes of each component is zero or approximately zero over the entire sequence; each component satisfies the condition that its local mean is zero.
[0031] The results of multivariate empirical mode decomposition are as follows Figure 3 and Figure 4 As shown.
[0032] S4: Construct the MGSNet model, preprocess the intrinsic mode functions (IMFs) and residuals obtained from the decomposition, and input them into the model for training and prediction to obtain the prediction results.
[0033] In step S4, the MGSNet model contains a gated recurrent unit, which uses two gate structures to update and transmit information. The update gate is responsible for controlling the influence of the previous state information on the current state, and the reset gate is responsible for controlling the ignoring of the previous state information.
[0034] The expression for the information state transfer at time t in the gated loop unit structure is: ; ; ; ; In the formula, This is represented by the processing formula for the update gate. This is represented as the processing formula for the reset gate. This is represented as the unit state candidate vector at the current time. Represented as output value, and This represents the weight and bias corresponding to each gate. Represented as the sigmoid function, Represented as element-wise product Each decomposition component After normalization, the normalized values are used as input to the MGSNet model for training and prediction, yielding normalized predicted values for each decomposed component. The calculation formula is: ; In the formula, Indicates the first Data for each decomposed component Represents the normalized i-th Each decomposed component data, and They represent the first The maximum and minimum values of each decomposed component data.
[0035] S5: Perform inverse normalization and reconstruction on the prediction results to obtain the effective wave height prediction value.
[0036] The prediction results are inversely normalized, and the calculation formula is as follows: ; In the formula, and They represent the first The maximum and minimum values of each decomposed component data Indicates the first Normalized predicted values of each decomposed component, Indicates the i-th after denormalization Predicted values for each decomposed component.
[0037] The predicted values of each decomposition component are reconstructed to obtain the reconstructed effective wave height prediction value in the normalized space. The calculation formula is: ; In the formula, This represents the reconstructed predicted effective wave height in the normalized space. This represents the inverse normalized prediction value of the k-th decomposed variable at time t.
[0038] The reconstruction is performed by reconstructing the predicted effective wave height in the normalized space, and the calculation formula is as follows: ; In the formula, and These represent the maximum and minimum values in the effective wave height time series data before normalization, respectively. This represents the sum of the predicted values of each decomposition component after inverse normalization.
[0039] To verify the effectiveness of the short-term significant wave height prediction method based on MGSNet provided in this invention, this invention uses the MGSNet model to predict the significant wave height for the next 1 hour and compares it with the actual values observed by buoys over time. Figure 6 As shown. To facilitate observation of the prediction effect, further comparison results of local time series are given, such as... Figure 7 As shown.
[0040] To further verify the effectiveness of this method, the performance of the proposed method is compared with that of BP, MLP, RF, and GRU methods in predicting the significant wave height in the next 1 hour. Model performance is evaluated using the root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (CC). RMSE and MAE measure the magnitude of the error between the predicted and actual values; smaller values indicate smaller prediction errors and higher prediction accuracy. CC measures the linear correlation between the predicted and actual sequences; values closer to 1 indicate better fit. The performance metrics are calculated using the following formulas: ; ; ; In the formula, and They represent variables respectively. and No. Data from one sample, and They represent variables respectively. and The average value, Indicates the number of samples.
[0041] The results of the method comparison are shown in Table 1: Table 1
[0042] When predicting the significant wave height in the next 1 hour using the MGSNet model of this invention, the RMSE, MAE, and CC are 0.061, 0.049, and 0.997, respectively; for GRU, the RMSE, MAE, and CC are 0.128, 0.096, and 0.980; for RF, the RMSE, MAE, and CC are 0.122, 0.088, and 0.979; for MLP, the RMSE, MAE, and CC are 0.146, 0.122, and 0.979; and for BP, the RMSE, MAE, and CC are 0.161, 0.120, and 0.963. Therefore, on the dataset of this embodiment, the method of this invention exhibits superior predictive performance in terms of RMSE, MAE, and CC.
[0043] Experimental results show that the present invention can effectively characterize the complex relationships and time dependencies among input variables under multivariate input conditions, and has good predictive ability for short-term changes in significant wave height, proving the effectiveness of the improved method presented in this paper. By combining the advantages of existing algorithms, the MGSNet model improves the prediction accuracy of short-term significant wave height, providing reliable technical support for practical applications.
[0044] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be the technical solution described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.
Claims
1. A method for predicting the significant wave height of short-term ocean waves based on MGSNet, characterized in that, Includes the following steps: S1: Acquire effective wave height data and marine environmental data using buoys, and perform preprocessing; S2: The correlation coefficient method is used to calculate the correlation coefficient between marine environmental data and significant wave height, and the time series dataset is obtained after screening and preprocessing. S3: Perform multivariate empirical mode decomposition (MEMD) on the time series dataset to obtain intrinsic mode functions (IMF) and residuals; S4: Construct the MGSNet model, preprocess the intrinsic mode functions (IMF) and residuals obtained from the decomposition, and input them into the model for training and prediction to obtain the prediction results; S5: Perform inverse normalization and reconstruction on the prediction results to obtain the effective wave height prediction value.
2. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, In step S1, the preprocessing includes performing time resolution unification processing on the acquired valid wave height data and marine environment data, and performing outlier removal and time alignment processing on the data.
3. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, In step S2, the correlation coefficient between marine environmental data and significant wave height is calculated using the correlation coefficient method. The calculation formula is as follows: ； In the formula, E( ) represents the expected value, S represents the time series of the significant wave height, and O represents the time series of any input variable.
4. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, Step S3 includes the following sub-steps: 1) Perform Multivariate Empirical Mode Decomposition (MEMD) on the time series dataset to obtain a multivariate time series. The expression for the multivariate time series is: ； In the formula, Indicates the first A marine environmental variable at time Time series data, Indicates the number of variables; 2) Decompose the multivariate time series X(t) into several intrinsic mode functions and residuals, expressed as: ； In the formula, Indicates the first Eigenmode function vector, This indicates the number of IMFs obtained from the decomposition. Represents the residual.
5. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 4, characterized in that, The intrinsic mode function has the following characteristics: the difference between the number of local extrema and the number of zero crossings of each component does not exceed 1; the local mean of the upper and lower envelopes of each component is zero or approximately zero over the entire sequence; and each component satisfies the condition that its local mean is zero.
6. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, In step S4, the MGSNet model includes a gated loop unit, which uses two gate structures to update and transmit information; the update gate is responsible for controlling the influence of the previous state information on the current state, and the reset gate is responsible for controlling the ignoring of the previous state information.
7. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, In step S5, the prediction result is inversely normalized, and the calculation formula is: ； In the formula, and They represent the first The maximum and minimum values of each decomposed component data Indicates the first Normalized predicted values of each decomposed component, Indicates the i-th after denormalization Predicted values for each decomposed component.
8. The method for predicting the significant wave height of short-term ocean waves based on MGSNet according to claim 1, characterized in that, In step S5, the reconstruction involves reconstructing the predicted effective wave height in the normalized space, calculated as follows: ； In the formula, and These represent the maximum and minimum values in the effective wave height time series data before normalization, respectively. This represents the sum of the predicted values of each decomposition component after inverse normalization.