Wave height short-term prediction system and method based on variational mode decomposition and wind and wave time correlation

Abstract

The invention discloses a wave height short-term prediction system and method based on variational mode decomposition and wind and wave time correlation, and belongs to the technical field of ocean engineering.The method comprises the steps that firstly, original storm time series data are decomposed into IMF components with different center frequencies through variational mode decomposition; secondly, performing data cleaning on the decomposed IMF component based on a Pearson's correlation coefficient; then combining two wind speed IMF components with the strongest correlation with the wave height IMF component based on correlation analysis to form a wind speed wave height component combination, and determining an input time length through an autocorrelation coefficient ACF of wave height data and a cross correlation coefficient CCF of the wind speed IMF components and the wave height IMF components; and finally, according to the formed wind speed wave height component combination and the determined input time length, carrying out short-term prediction on the wave height based on a long short-term memory neural network. According to the system and the method provided by the invention, the influence of noise on prediction is reduced, and the prediction precision is improved while the prediction efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean engineering, and particularly relates to a short-term wave height prediction system and method based on variational mode decomposition and the time correlation of wind and waves. Background Technique

[0002] Waves, as a key element of ocean meteorology, have an important impact on maritime navigation, construction activities, ocean energy development, etc. Therefore, accurate wave prediction is crucial for related fields. Currently, wave prediction mainly relies on physical models such as SWAN, WAM, and Wave Watch III. These numerical models perform forecasts by simulating the propagation and development process of waves. However, traditional physical models have limitations in calculation accuracy and efficiency, especially in short-term wave prediction, and it is difficult to cope with the complexity of environmental changes.

[0003] In recent years, with the development of deep learning technology, scholars have attempted to apply machine learning methods to wave prediction, especially by combining data from monitoring devices such as hydrometeorological stations and offshore buoys. However, existing deep learning models usually perform predictions by directly fusing wind speed and wave height signals, but do not fully consider the mutual influence between different frequency components in the signals. Since the signals of wind speed and wave height have obvious period differences, this direct fusion method may cause signals with low correlation to interfere with model learning, thereby affecting the prediction performance. Therefore, there are still deficiencies in the existing research on the fusion method of wind and wave data, especially in how to handle the time correlation of wind and wave signals.

[0004] For this reason, this paper proposes a new wave height prediction method, which extracts different frequency components of wind and wave signals through variational mode decomposition (VMD) and performs data fusion based on their time correlation, so as to improve the accuracy of short-term wave height prediction. Summary of the Invention

[0005] The purpose of the present invention is to provide a short-term wave height prediction system and method based on variational mode decomposition and the time correlation of wind and waves, which perform noise reduction and feature extraction on wind and wave data through variational mode decomposition, fully consider the time correlation of wind and wave characteristics, and realize the fusion of wind and wave data through a long short-term memory neural network to achieve the purpose of short-term wave height prediction.

[0006] To achieve the above purpose, the present invention provides a short-term wave height prediction system based on variational mode decomposition and the time correlation of wind and waves, including a data decomposition module, a data cleaning module, a correlation analysis module, and a prediction module;

[0007] The data decomposition module decomposes the original wind-wave time series data into IMF components (intrinsic mode components) with different central frequencies through variational mode decomposition. The IMF components include the wind speed IMF component and the wave height IMF component. Among them, the number of decompositions and the penalty factor parameter of variational mode decomposition are determined according to the frequency domain information of the data;

[0008] The data cleaning module calculates the Pearson correlation coefficient between each IMF component and the original wind-wave time series data, identifies the IMF components with low Pearson correlation coefficients as noise and removes them to achieve data cleaning;

[0009] The correlation analysis module first calculates the Pearson correlation coefficient between each wind speed IMF component and wave height IMF component, combines the two wind speed IMF components with the strongest correlation with each wave height IMF component to form a wind speed-wave height component combination for subsequent wave height prediction. Subsequently, by calculating the changes of the autocorrelation coefficient and cross-correlation coefficient between each wind speed IMF component and wave height IMF component with the lag time, the appropriate input time length is determined to ensure that sufficient information can be obtained for subsequent wave height prediction while taking into account the prediction efficiency;

[0010] The prediction module inputs the historical data of the wind speed-wave height component combination with strong correlation into the long short-term memory neural network and uses the wind-wave data fusion to achieve short-term prediction of the wave height.

[0011] The present invention also provides a method for a short-term wave height prediction system based on variational mode decomposition and the time correlation between wind and waves, including the following steps:

[0012] Step 1: Decompose the original wind-wave time series data into IMF components with different central frequencies through variational mode decomposition; the original wind-wave time series data includes the original wind speed time series data and the original wave height time series data;

[0013] Step 2: Perform data cleaning on the decomposed IMF components based on the Pearson correlation coefficient;

[0014] Step 3: Combine the two wind speed IMF components with the strongest correlation with the wave height IMF component based on correlation analysis to form a wind speed-wave height component combination Comb i , i = 1, 2, 3,..., m, and determine the input time length through the autocorrelation coefficient ACF of the wave height data and the cross-correlation coefficient CCF between the wind speed IMF component and the wave height IMF component;

[0015] Step 4: Based on the wind speed-wave height component combination formed in Step 3 and the determined input time length, perform short-term prediction of the wave height based on the long short-term memory neural network.

[0016] Preferably, in step 1, the process of decomposing the original wind-wave time series data into IMF components with different central frequencies by variational mode decomposition is as follows:

[0017] S11. Process the time series data of wind speed and wave height through fast Fourier transform to obtain the frequency domain information of the time series data of wind speed and wave height. Determine the number of mode decompositions and the penalty factor during variational mode decomposition according to the frequency domain information. The calculation expressions are as follows:

[0018]

[0019] where, x n is the nth sample of the time domain signal, X k is the kth frequency component of the frequency domain signal, and N is the length of the time series signal;

[0020] S12. According to the determined number of mode decompositions and the penalty factor, decompose the time series data of wind speed and wave height into IMF components with different central frequencies by variational mode decomposition, and arrange each IMF component in ascending order of the value of the central frequency; among them, the expression of the constrained variational model of variational mode decomposition is as follows:

[0021]

[0022] In the formula, is the single-sided spectrum of each IMF component, represents modulating the spectrum of each IMF component to the corresponding base frequency band, u k (t) represents the decomposed IMF component, ω k represents the central frequency of each component, represents the differential in the time domain, δ(t) represents the Dirac function, x(t) represents the original time series data, and j represents the imaginary unit.

[0023] Preferably, in step 2, the process of data cleaning for the decomposed IMF components based on the Pearson correlation coefficient is as follows: Calculate the Pearson correlation coefficients of each IMF component with the original wind speed time series data and the original wave height time series data respectively, and then identify and remove the IMF components with low correlation coefficients; the calculation expressions are as follows:

[0024]

[0025] In the formula, z i is the original data, is the value of the IMF component, and are the average values of the original data and the IMF component respectively, and R 2 represents the Pearson correlation coefficient.

[0026] Preferably, in step 3, the calculation expressions of the autocorrelation coefficient ACF of the wave height data and the cross-correlation coefficient CCF of the wind speed IMF component and the wave height IMF component are as follows:

[0027]

[0028]

[0029] In the formula, k represents the number of lag time steps, x t represents the value of the wave height at time t, x t-k represents the value of the wave height at time t-k, y t-k represents the value of the wind speed at time t-k, represents the average value of the wave height time series data, represents the average value of the wind speed time series data.

[0030] Preferably, in step 4, based on the wind speed-wave height component combination formed in step 3 and the determined input time length, the process of predicting the wave height in the short term based on the long short-term memory neural network is as follows:

[0031] S41. Use the long short-term memory neural network to predict the wave height for each wind speed-wave height component combination respectively, and obtain the predicted value Pred i of each wave height component, and the expression is as follows:

[0032] Pred i = LSTM{Comb i}, i = 1, 2, 3, … m (6)

[0033] S42. Sum up the predicted values of each wave height component to finally obtain the short-term predicted value FinalPred of the wave height, and the expression is as follows:

[0034]

[0035] Therefore, the wave height short-term prediction system and method based on variational mode decomposition and the time correlation of wind and waves of the present invention have the following beneficial effects:

[0036] (1) Improve data quality: The data decomposition module decomposes the original data into IMF components with different center frequencies through variational mode decomposition technology, and the data cleaning module removes the IMF components (noise) with low correlation coefficients, thereby improving the data quality for prediction;

[0037] (2) Optimize data combination: The correlation analysis module constructs the wind speed-wave height component combination by calculating the correlation coefficient, making the data combination participating in the prediction more reasonable and better reflecting the relationship between wind and waves, which helps to improve the prediction accuracy;

[0038] (3) Improve prediction efficiency: The correlation analysis module determines the appropriate input time length, taking into account both the prediction efficiency while ensuring sufficient information for wave height prediction, and avoiding unnecessary waste of computing resources.

[0039] (4) Enhance prediction accuracy: The overall method effectively enhances the accuracy of short-term wave height prediction by fully considering the time correlation of wind and wave characteristics and combining long short-term memory neural networks for short-term wave height prediction, compared with existing methods that do not fully consider the mutual influence of frequency components and time correlation.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0041] Figure 1 is a structural schematic diagram of a short-term wave height prediction system based on variational mode decomposition and the time correlation of wind and waves according to the present invention;

[0042] Figure 2 is an overall process block diagram of the method of a short-term wave height prediction system based on variational mode decomposition and the time correlation of wind and waves according to the present invention;

[0043] Figure 3 is a variational mode decomposition process diagram of an embodiment of the present invention;

[0044] Figure 4 is a data cleaning process diagram of an embodiment of the present invention;

[0045] Figure 5 is a process diagram of the formation of the wind speed-wave height component combination of an embodiment of the present invention;

[0046] Figure 6 is a process diagram of predicting the short-term wave height of an embodiment of the present invention. Detailed Embodiments

[0047] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0048] Please refer to Figure 1 , a short-term wave height prediction system based on variational mode decomposition and the time correlation of wind and waves, includes a data decomposition module, a data cleaning module, a correlation analysis module, and a prediction module;

[0049] The data decomposition module decomposes the original wind-wave time series data into IMF components (intrinsic mode components) with different central frequencies through variational mode decomposition. Each IMF component will be arranged and numbered in ascending order of the central frequency. The specific decomposition is as shown in Figure 3 The IMF components include the wind speed IMF component and the wave height IMF component. Among them, the number of decomposition and the penalty factor parameter of variational mode decomposition are determined according to the frequency domain information of the data.

[0050] The data cleaning module calculates the Pearson correlation coefficient between each IMF component and the original wind-wave time series data, identifies the IMF components with low Pearson correlation coefficients as noise and removes them to achieve data cleaning. The specific cleaning process is as shown in Figure 4 The IMF components corresponding to the noise to be removed after identification are shown in the dotted box.

[0051] As shown in Figure 5 The correlation analysis module first calculates the Pearson correlation coefficient between each wind speed IMF component and wave height IMF component, combines the two wind speed IMF components with the strongest correlation with each wave height IMF component to form a wind speed-wave height component combination for subsequent wave height prediction. Subsequently, by calculating the changes of the autocorrelation coefficient and cross-correlation coefficient between each wind speed IMF component and wave height IMF component with the lag time, the appropriate input time length is determined to ensure that sufficient information can be obtained for subsequent wave height prediction while taking into account the prediction efficiency.

[0052] As shown in Figure 6 The prediction module inputs the historical data of the strongly correlated wind speed-wave height component combination into a long short-term memory neural network, and uses wind-wave data fusion to achieve short-term wave height prediction, thereby obtaining the prediction model corresponding to each wind speed-wave height component combination. The input data lengths of each prediction model are different and all come from the input lengths obtained in the correlation analysis module. Subsequently, the prediction model completes the prediction to obtain the wave height prediction values of each wind speed-wave height component combination. Finally, the prediction values of each prediction model are added together to obtain the final wave height prediction value.

[0053] Please refer to Figure 2 A method for a short-term wave height prediction system based on variational mode decomposition and the time correlation of wind and waves includes the following steps:

[0054] Step 1: Decompose the original wind-wave time series data into IMF components with different central frequencies through variational mode decomposition. The original wind-wave time series data includes the original wind speed time series data and the original wave height time series data. Among them, the process of decomposing the original wind-wave time series data into IMF components with different central frequencies through variational mode decomposition is as follows:

[0055] S11. Process the time series data of wind speed and wave height through fast Fourier transform to obtain the frequency domain information of the time series data of wind speed and wave height. Determine the number of mode decompositions and the penalty factor during variational mode decomposition according to the frequency domain information. The calculation expressions are as follows:

[0056]

[0057] where x n is the nth sample of the time domain signal, X k is the kth frequency component of the frequency domain signal, and N is the length of the time series signal;

[0058] S12. According to the determined number of mode decompositions and the penalty factor, decompose the time series data of wind speed and wave height into IMF components with different center frequencies through variational mode decomposition, and arrange each IMF component in ascending order of the center frequency value. Among them, the expression of the constrained variational model of variational mode decomposition is as follows:

[0059]

[0060] In the formula, is the single-sided spectrum of each IMF component, represents modulating the spectrum of each IMF component to the corresponding base frequency band, u k (t) represents the decomposed IMF component, ω k represents the center frequency of each component, represents the differentiation in the time domain, δ(t) represents the Dirac function (used to ensure the locality of the signal), x(t) represents the original time series data, and j represents the imaginary unit.

[0061] Step 2. Perform data cleaning on the decomposed IMF components based on the Pearson correlation coefficient; the specific process is as follows: Calculate the Pearson correlation coefficients of each IMF component with the original wind speed time series data and the original wave height time series data respectively, and then identify and remove the IMF components with low correlation coefficients. The calculation expressions are as follows:

[0062]

[0063] In the formula, z i is the original data, is the value of the IMF component, and are the average values of the original data and the IMF component respectively, and R 2 represents the Pearson correlation coefficient.

[0064] Step 3: Combine the two wind speed IMF components with the strongest correlation with the wave height IMF component based on correlation analysis. Specifically: Calculate the Pearson correlation coefficient between each wind speed IMF component and each wave height IMF component. The calculation expression is as shown in Equation (3). Combine the two wind speed IMF components with the strongest correlation with each wave height IMF component to form the wind speed - wave height component combination Comb i , i = 1, 2, 3, …, m, and determine the input time length through the autocorrelation coefficient ACF of the wave height data and the cross - correlation coefficient CCF between the wind speed IMF component and the wave height IMF component. The calculation expressions for the autocorrelation coefficient ACF of the wave height data and the cross - correlation coefficient CCF between the wind speed IMF component and the wave height IMF component are as follows:

[0065]

[0066]

[0067] In the formula, k represents the number of lag time steps, x t represents the value of the wave height at time t, x t-k represents the value of the wave height at time t - k, y t-k represents the value of the wind speed at time t - k, represents the average value of the wave height time - series data, represents the average value of the wind speed time - series data.

[0068] Step 4: Based on the wind speed - wave height component combination formed in Step 3 and the determined input time length, perform short - term prediction of the wave height based on the long short - term memory neural network. The specific process is as follows:

[0069] S41: Use the long short - term memory neural network to perform wave height prediction for each wind speed - wave height component combination respectively, and obtain the predicted value Pred i of each wave height component. The expression is as follows:

[0070] Pred i = LSTM{Comb i}, i = 1, 2, 3, …m(6)

[0071] S42: Sum up the predicted values of each wave height component to finally obtain the short - term predicted value FinalPred of the wave height. The expression is as follows:

[0072]

[0073] Therefore, the present invention adopts the above short-term wave height prediction system and method based on variational mode decomposition and the time correlation of wind and waves. The variational mode decomposition is used to denoise and extract features from wind and wave data. At the same time, the time correlation of wind and wave features is fully considered, and the long short-term memory neural network is used to realize the fusion of wind and wave data, so as to achieve the purpose of short-term wave height prediction.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation, characterized by: It includes data decomposition module, data cleaning module, correlation analysis module and prediction module; The data decomposition module decomposes the original wind and wave time series data into IMF components with different central frequencies through variational mode decomposition. The IMF components include wind speed IMF components and wave height IMF components. The number of decompositions and penalty factor parameters of variational mode decomposition are determined according to the frequency domain information of the data; The data cleaning module calculates the Pearson correlation coefficient between each IMF component and the original wind and wave time series data, identifies the IMF components with low Pearson correlation coefficient as noise and removes them to achieve data cleaning; The correlation analysis module first calculates the Pearson correlation coefficient between each wind speed IMF component and the wave height IMF component, and combines the two wind speed IMF components with the strongest correlation with each wave height IMF component to form a wind speed wave height component combination; then, the input time length is determined by calculating the changes in the autocorrelation coefficient and the mutual correlation coefficient between each wind speed IMF component and the wave height IMF component with the lag time; The prediction module inputs the historical data of the strongly correlated wind speed and wave height components into the long short-term memory neural network, and uses the fusion of wind and wave data to achieve short-term prediction of wave height.

2. A method for a short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation as claimed in claim 1, characterized in that: The following steps are involved: Step 1: Decompose the original wind and wave time series data into IMF components with different central frequencies by variational mode decomposition; the original wind and wave time series data includes original wind speed time series data and original wave height time series data; Step 2: Clean the decomposed IMF components based on the Pearson correlation coefficient; Step 3: Based on the correlation analysis, the two wind speed IMF components with the strongest correlation with the wave height IMF component are combined to form the wind speed wave height component combination Comb i , i=1,2,3,…,m, and the input time length is determined by the autocorrelation coefficient ACF of the wave height time series data and the cross-correlation coefficient CCF between the wind speed IMF component and the wave height IMF component; Step 4: According to the wind speed wave height component combination formed in step 3 and the determined input time length, the wave height is predicted in the short term based on the long short-term memory neural network.

3. The method of the short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation according to claim 2 is characterized in that: The process of decomposing the original wind and wave time series data into IMF components with different center frequencies by variational mode decomposition in step 1 is as follows: S11. The time series data of wind speed and wave height are processed by fast Fourier transform to obtain the frequency domain information of the time series data of wind speed and wave height. The number of modal decompositions and penalty factors during variational mode decomposition are determined according to the frequency domain information. The calculation expression is as follows: Among them, x n is the nth sample of the time domain signal, X k is the kth frequency component of the frequency domain signal, and N is the length of the time series signal; S12. According to the determined number of modal decompositions and penalty factors, the time series data of wind speed and wave height are decomposed into IMF components with different center frequencies through variational modal decomposition, and the IMF components are arranged from low to high according to the value of the center frequency; wherein, the expression of the constrained variational model of variational modal decomposition is as follows: In the formula, is the one-sided spectrum of each IMF component, Indicates that the spectrum of each IMF component is modulated to the corresponding baseband, u k (t) represents the decomposed IMF component, ω k represents the center frequency of each component, represents the differential in the time domain, δ(t) represents the Dirac function (used to ensure the locality of the signal), x(t) represents the original time series data, and j represents the imaginary unit.

4. The method of the short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation according to claim 3 is characterized in that: The process of data cleaning of the decomposed IMF components based on the Pearson correlation coefficient in step 2 is: calculate the Pearson correlation coefficient of each IMF component with the original wind speed time series data and the original wave height time series data respectively, and then identify the IMF components with low correlation coefficients as noise and remove them; the calculation expression is as follows: In the formula, z i is the original data, is the value of the IMF component, and are the original data average and the IMF component average, R 2 represents the Pearson correlation coefficient.

5. The method of the short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation according to claim 4 is characterized in that: In step 3, the calculation expressions of the autocorrelation coefficient ACF of wave height data and the cross-correlation coefficient CCF between the wind speed IMF component and the wave height IMF component are as follows: In the formula, k represents the number of lag time steps, x t represents the value of the wave height at time t, x t-k Indicates the value of the wave height at time tk, y t-k represents the value of wind speed at time tk, represents the average value of the wave height time series data, Represents the average value of wind speed time series data.

6. The method of the short-term wave height prediction system based on variational mode decomposition and wind and wave time correlation according to claim 5 is characterized in that: In step 4, according to the combination of wind speed and wave height components formed in step 3 and the determined input time length, the process of short-term prediction of wave height based on long short-term memory neural network is as follows: S41. Use the long short-term memory neural network to predict the wave height of each wind speed wave height component combination, and obtain the predicted value Pred of each wave height component i , the expression is as follows: Pred i =LSTM{Comb i },i=1,2,3,…m(6) S42, add up the predicted values ​​of each wave height component, and finally obtain the short-term predicted value of the wave height FinalPred, which is expressed as follows:

Images