Multi-factor ocean wave forecasting method based on signal decomposition, electronic equipment, and medium

Abstract

The present invention discloses a multi-factor ocean wave forecasting method based on signal decomposition, an electronic device, and a medium, comprising: performing adaptive noise complete empirical mode decomposition on effective wave height, effective wave period, and wave direction time series data respectively, dividing each intrinsic mode function obtained by decomposition into a high-frequency mode function and a low-frequency mode function; dividing the low-frequency mode function corresponding to the effective wave height, effective wave period, and wave direction time series data into K 11 , K 21 , K 31 The high-frequency modal functions corresponding to the effective wave height, effective wave period, and wave direction time series data are superimposed and then subjected to variational modal decomposition. The decomposed modal functions are divided into K 12 , K 22 , K 32 Group mode; K 11 , K 12 The group modes are input into the BiLSTM model in sequence, and the effective wave height prediction results are output; 21 , K 22 The group modes are input into the LSTM model in sequence, and the prediction results of the effective wave period are output; 31 , K 32 The group modes are sequentially input into the LSTM‑GRU model, and the wave direction prediction results are output.

Description

Technical Field

[0001] The invention belongs to the field of marine science, and in particular relates to a multi-factor ocean wave forecasting method based on signal decomposition, electronic equipment, and medium. Background Art

[0002] At present, the wave prediction method based on deep learning has the following problems:

[0003] Most methods only focus on the prediction of significant wave height, and no one has ever paid attention to the performance differences of different neural networks in predicting different wave characteristics; however, in practical applications, not only significant wave height, but also significant wave period and wave direction are important physical quantities with predictive value, and different neural network architectures have considerable differences in the prediction accuracy of different wave characteristics.

[0004] The wave prediction method based on deep learning has low accuracy in long-term prediction and generalized prediction, and it is difficult to effectively improve the prediction accuracy simply by changing the neural network architecture.

[0005] At the same time, the current signal decomposition technology, mainly EMD (Empirical Mode Decomposition), faces the following limitations in the field of deep learning wave prediction:

[0006] Although the EMD signal decomposition method can adaptively decompose time series, the number of IMFs decomposed will change with the length and complexity of the time series. As a result, the model trained based on each IMF does not correspond to the number and physical meaning of each IMF to be predicted during generalization, making it impossible to predict.

[0007] Even if VMD with adjustable decomposition number is used to obtain the same number of modal functions, the decomposed modal functions still cannot physically correspond to the modal functions to be predicted during generalization, resulting in a significant decrease in prediction accuracy. Summary of the invention

[0008] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a multi-factor ocean wave forecasting method based on signal decomposition, an electronic device and a medium.

[0009] In a first aspect, an embodiment of the present invention provides a multi-factor ocean wave forecasting method based on signal decomposition, the method comprising:

[0010] Acquire ocean wave data, and extract effective wave height time series data, effective wave period time series data, and wave direction time series data from the ocean wave data;

[0011] Perform adaptive noise complete empirical mode decomposition on the effective wave height time series data, the effective wave period time series data, and the wave direction time series data, and calculate the entropy value of each intrinsic mode function obtained by decomposition; set the entropy threshold, and take the intrinsic mode function greater than the entropy threshold as the high-frequency mode function, and take the intrinsic mode function less than the entropy threshold as the low-frequency mode function;

[0012] The low-frequency modal functions corresponding to the effective wave height time series data, effective wave period time series data, and wave direction time series data are divided into K 11 Group mode, K 21 Group mode, K 31 Group modality;

[0013] The high-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data are superimposed and then subjected to variational modal decomposition. The modal functions obtained after variational modal decomposition are divided into K 12 Group mode, K 22 Group mode, K 32 Group modality;

[0014] K 11 Group mode, K 12 The group modes are input into the BiLSTM model in sequence, and the effective wave height prediction results are output; 21 Group mode, K 22 The group modes are input into the LSTM model in sequence, and the prediction results of the effective wave period are output; K 31 Group mode, K 32 The group modes are input into the LSTM-GRU model in sequence, and the wave direction prediction results are output.

[0015] In a second aspect, an embodiment of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned multi-factor wave forecasting method based on signal decomposition.

[0016] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned multi-factor wave forecasting method based on signal decomposition.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the above-mentioned multi-factor wave forecasting method based on signal decomposition.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] (1) The present invention performs adaptive noise complete empirical mode decomposition (ICEEMDAN decomposition) and variational mode decomposition (VMD decomposition) on the significant wave height time series data, the significant wave period time series data, and the wave direction time series data, so that the neural network can better learn the information in the time series of each wave characteristic, thereby significantly improving the prediction accuracy.

[0020] (2) The present invention uses the RMSE-Ci classification technology to divide the low-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data into K 11 Group mode, K 21 Group mode, K 31 The modal functions obtained after the variational modal decomposition of the high-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data are divided into K 12 Group mode, K 22 Group mode, K 32 Group mode; solve the difficulty that the current EMD-type signal decomposition method cannot be generalized when combined with deep learning, so that the model can be generalized while retaining the improvement of prediction accuracy by signal decomposition technology.

[0021] (3) The present invention uses the BiLSTM model to predict the effective wave height, the LSTM model to predict the effective wave period, and the LSTM-GRU model to predict the wave direction. The neural network architecture with the best prediction performance is selected for the effective wave height, effective wave period, and wave direction to achieve high-precision long-term generalizable prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0023] Figure 1 A flow chart of a multi-factor ocean wave forecasting method based on signal decomposition provided by an embodiment of the present invention;

[0024] Figure 2 A schematic block diagram of a multi-factor ocean wave forecasting method based on signal decomposition provided by an embodiment of the present invention;

[0025] Figure 3 A generalized prediction result diagram provided by an embodiment of the present invention;

[0026] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] It should be noted that, in the absence of conflict, the features in the following embodiments and implementations may be combined with each other.

[0029] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a multi-factor ocean wave forecasting method based on signal decomposition, the method comprising:

[0030] Step S1, acquiring ocean wave data, and decomposing the ocean wave data into effective wave height time series data, effective wave period time series data, and wave direction time series data.

[0031] Step S2, perform adaptive noise complete empirical mode decomposition on the effective wave height time series data, the effective wave period time series data, and the wave direction time series data, respectively, and calculate the entropy value of each intrinsic mode function obtained by the decomposition; set the entropy threshold, and take the intrinsic mode function with an entropy value greater than the entropy threshold as the high-frequency mode function, and take the intrinsic mode function with an entropy value less than the entropy threshold as the low-frequency mode function.

[0032] Specifically, Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) is performed on the effective wave height time series data to obtain several first intrinsic mode functions; Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) is performed on the effective wave period time series data to obtain several second intrinsic mode functions; Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) is performed on the wave direction time series data to obtain several third intrinsic mode functions;

[0033] Calculate the entropy value corresponding to each first eigenmode function, second eigenmode function, and third eigenmode function, set an entropy threshold, and use the eigenmode function with an entropy value less than the entropy threshold as a low-frequency mode function. In this example, the entropy threshold is set to 4.

[0034] Furthermore, the process of calculating the entropy value corresponding to each first eigenmode function, second eigenmode function, and third eigenmode function includes:

[0035] The intrinsic mode function is processed by a low-pass Butterworth filter to obtain sub-time series of each intrinsic mode function at several scales; in this example, the scale (Scales) of the low-pass Butterworth filter is 5, the order is 3, the cutoff frequency factor is 0.6, and the cutoff frequency is equal to the cutoff frequency factor divided by the current scale.

[0036] The sample entropy (SampleEntropy) of each sub-time series of each intrinsic mode function at each scale is calculated, and the sum of the sample entropies at each scale is taken as the entropy value of the intrinsic mode function.

[0037] Step S3: divide the low-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data into K 11 Group mode, K 21 Group mode, K 31 Group modality.

[0038] Furthermore, the low-frequency modal functions corresponding to the effective wave height time series data are divided into K 11 The low-frequency modal function corresponding to the effective wave period time series data is divided into K 21 Group mode, divide the low-frequency mode function corresponding to the wave direction time series data into K 31 Group modality.

[0039] Step S4, superimposing the high-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data, respectively, and performing variational modal decomposition; the modal functions obtained after the variational modal decomposition are divided into K 12 Group mode, K 22 Group mode, K 32 Group modality.

[0040] Specifically, the high-frequency modal functions corresponding to the significant wave height time series data are superimposed, and several first modal functions are obtained after variational mode decomposition (VMD). The first modal function is divided into K 12 Group modality;

[0041] The high-frequency modal functions corresponding to the effective wave period time series data are superimposed, and several second modal functions are obtained after variational mode decomposition; the second modal function is divided into K 22 Group modality;

[0042] The high-frequency modal functions corresponding to the wave direction time series data are superimposed, and several third modal functions are obtained after variational mode decomposition; the third modal function is divided into K 32 Group modality.

[0043] Exemplarily, in this example, the low-frequency modal function corresponding to the effective wave height time series data is divided into three groups, including: intermediate frequency IMF (>1.40), low frequency IMF (<1.40 and >1.39), and trend item IMF (<1.39); the first modal function corresponding to the effective wave height time series data is divided into three groups, including: high frequency mode (>9.1), intermediate frequency mode (<9.1 and >7.4), and low frequency mode (<7.4).

[0044] The low-frequency modal functions corresponding to the effective wave period time series data are divided into three groups, including: medium-frequency IMF (>1.20), low-frequency IMF (<1.20 and >0.43), and trend item IMF (<0.43); the second modal functions corresponding to the effective wave period time series data are divided into three groups, including: high-frequency mode (>11.9), medium-frequency mode (<11.9 and >8.0), and low-frequency mode (<8.0).

[0045] The low-frequency modal functions corresponding to the wave direction time series data are divided into three groups, including: intermediate frequency IMF (>1.80), low frequency IMF (<1.80 and >1.67), and trend item IMF (<1.67); the third modal functions corresponding to the wave direction time series data are divided into three groups, including: high frequency mode (>9.2), intermediate frequency mode (<9.2 and >7.1), and low frequency mode (<7.1).

[0046] Step S5: K 11 Group mode, K 12 The group modes are input into the BiLSTM model in sequence, and the effective wave height prediction results are output; 21 Group mode, K 22 The group modes are input into the LSTM model in sequence, and the prediction results of the effective wave period are output; K 31 Group mode, K 32 The group modes are input into the LSTM-GRU model in sequence, and the wave direction prediction results are output.

[0047] Specifically, K 11 Group mode, K 12 The group modes are sequentially input into the pre-trained BiLSTM model, and the prediction sub-results corresponding to each group mode are added together as the significant wave height prediction result;

[0048] K 21 Group mode, K 22 The group modes are input into the pre-trained LSTM model in sequence, and the prediction sub-results corresponding to each group mode are added together as the prediction result of the effective wave period;

[0049] K 31 Group mode, K 32The group modalities are input into the pre-trained LSTM-GRU model in sequence, and the prediction sub-results corresponding to each group modality are added together as the wave direction prediction result.

[0050] Furthermore, the BiLSTM model includes an input layer (Input Layer), a first bidirectional long short-term memory network (BiLSTM Layer), a first random dropout layer (Dropout Layer), a second bidirectional long short-term memory network, a second random dropout layer, and a fully connected layer (i.e., output layer, Dense Layer) connected in sequence.

[0051] The LSTM model includes an input layer (Input Layer), a first long short-term memory network (LSTMLayer), a first random dropout layer (Dropout Layer), a second long short-term memory network, a second random dropout layer, and a fully connected layer (i.e., output layer, Dense Layer) connected in sequence.

[0052] The LSTM-GRU model includes an input layer (Input Layer), a first long short-term memory network (LSTM Layer), a first random dropout layer (Dropout Layer), a first gated recurrent unit (GRU Layer), a second random dropout layer, a second long short-term memory network, a third random dropout layer, and a fully connected layer (i.e., output layer, Dense Layer) connected in sequence.

[0053] Furthermore, the training process of the BiLSTM model, LSTM model, and LSTM-GRU model includes: setting the batch size to 64, the number of iterations to 300, the Dropout rate to 0.5, the hidden layer dimension of the LSTM model and LSTM-GRU model to 64, and the hidden layer dimension of the BiLSTM model to 128. In addition, the learning rate scheduler and early stopping mechanism are introduced. The learning rate scheduler is ReduceLROnPlateau, the optimizer is Adam (Adaptive Moment Estimation), and the initial learning rate parameter is set to 0.001. The specific learning rate update strategy is that the learning scheduler will wait for 20 Epochs. If the loss on the validation set does not decrease in these 20 Epochs, the learning rate will be reduced by 70%; if the loss on the validation set still does not decrease after waiting for 30 Epochs, the training will be terminated early, and the model parameters will be restored to the model parameters when the validation set loss value is the smallest.

[0054] Furthermore, in this example, the coefficient of determination (R Square, R 2 ) to evaluate the prediction accuracy, the expression is as follows:

[0055]

[0056] in, is the predicted value, is the true value, The closer the coefficient of determination is to 1, the better.

[0057] Example 1

[0058] In this example, the sea wave data of a certain ocean station is obtained, and the present invention provides a multi-factor sea wave forecasting method based on signal decomposition to further illustrate. For the convenience of explanation, this example takes the effective wave period data of several stations on a certain island as an example.

[0059] Step S1, obtain the effective wave cycle time series data from January 1, 2019 to December 31, 2022 for the training station. There are omissions and outliers in the original data. If a single point of data is missing, the data of the next acquisition time is directly transferred; if the missing data is multi-point continuous data, all feature data of the time period are directly deleted. After data processing, the length of the original data used for training is 18203.

[0060] Obtain effective wave period data from three stations used to test the generalization performance of the model. The three stations are located in the north, central, and south of an island. The data collection time for the central station is from March 1 to March 31, 2023, and the data collection time for the north and south stations is from September 1 to September 30, 2022.

[0061] Step S2, use ICEEMDAN to decompose the effective wave period time series data to obtain each intrinsic mode function (IMF). Use RMSE-Ci to calculate the entropy value of each IMF, where the time scale number is set to 5, the low-pass Butterworth filter order is set to 3, the cutoff frequency factor is set to 0.6, and the remaining parameters are set to the default values. After the calculation is completed, the IMFs with entropy values ​​greater than 4 are merged into high-frequency mode functions (Model) and input into VMD for decomposition. The decomposition results and non-high-frequency IMFs are divided into 6 groups according to the following ratios:

[0062] Medium frequency IMF (>1.20), low frequency IMF (<1.20 and >0.43), trend item IMF (<0.43), high frequency mode (>11.9), medium frequency mode (<11.9 and >8.0), low frequency mode (<8.0).

[0063] The supervised data set is constructed based on the 6 modal functions obtained from the training data, with the training set ratio being 80% and the validation set being 20%. The label window of the training set only moves 1 time step at a time, but the label window of the validation set moves the time step of the label window length (prediction length). All the modal functions obtained from the 3 test data are used as the test set.

[0064] Step S3, using the training set to train the LSTM network, to demonstrate the prediction effect, a model based on LSTM prediction without signal decomposition is established, and the same training, verification, and test data are used.

[0065] The batch size of LSTM is set to 64, the number of iterations is set to 300, the Dropout rate is set to 0.5, and the hidden layer dimension is set to 128. In addition, the learning rate scheduler and early stopping mechanism are introduced to restore the model parameters to the model parameters with the minimum loss value of the validation set when the training is terminated early. For the convenience of representation, the input data length: output data length of this example is 1:1, and only the data with the prediction steps of 1, 3, 6, and 12 are given, where each step corresponds to 1 hour of data.

[0066] In step S4, the validation set is input into the trained LSTM network to obtain the prediction result. In order to show the effect of improving the prediction accuracy, a deep learning model without the signal decomposition step but with the same other processing is set up as a control group for generalization prediction.

[0067] The determination coefficients of the method proposed in this example when the prediction steps are 1, 3, 6, and 12 are:

[0068] North: 0.8386, 0.9126, 0.8738, 0.8072; control group: -7.8382, 0.5055, 0.3737, 0.3812;

[0069] Central: 0.8020, 0.8824, 0.8210, 0.7386; Control group: -8.9520, 0.4905, 0.3175, 0.4132;

[0070] South: 0.8685, 0.8944, 0.8780, 0.8248; Control group: -10.3864, 0.6755, 0.5940, 0.5488;

[0071] Figure 3 (A) shows the fitting of the prediction results of the southern station under this method to the true value. Figure 3 (B) shows the fitting of the prediction results of the central station under this method to the true value. Figure 3(C) in the figure shows the fitting of the prediction results of the northern station under this method to the true value. It can be seen that the prediction results fit the true value well.

[0072] In summary, the present invention provides a multi-factor wave forecasting method based on signal decomposition, which performs adaptive noise complete empirical mode decomposition (ICEEMDAN decomposition) and variational mode decomposition (VMD decomposition) on the effective wave height time series data, the effective wave period time series data, and the wave direction time series data, so that the neural network can better learn the information in the time series of each wave characteristic, thereby significantly improving the prediction accuracy.

[0073] The present invention uses RMSE-Ci classification technology to divide the low-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data into K 11 Group mode, K 21 Group mode, K 31 The modal functions obtained after the variational modal decomposition of the high-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data are divided into K 12 Group mode, K 22 Group mode, K 32 Group mode; solve the difficulty that the current EMD-type signal decomposition method cannot be generalized when combined with deep learning, so that the model can be generalized while retaining the improvement of prediction accuracy by signal decomposition technology.

[0074] The present invention respectively uses the BiLSTM model to predict the effective wave height, uses the LSTM model to predict the effective wave period, and uses the LSTM-GRU model to predict the wave direction; for the effective wave height, effective wave period, and wave direction, a neural network architecture with the best corresponding prediction performance is selected to achieve high-precision long-term generalizable prediction.

[0075] Accordingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned multi-factor wave forecasting method based on signal decomposition. Figure 4 As shown in FIG. 1 , a hardware structure diagram of any device with data processing capability for the multi-factor ocean wave forecasting method based on signal decomposition provided by an embodiment of the present invention is shown in FIG. Figure 4 In addition to the processor, memory and network interface shown, any device with data processing capability in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capability, which will not be described in detail.

[0076] Accordingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by the processor, the multi-factor wave forecasting method based on signal decomposition as described above is implemented. The computer-readable storage medium can be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium can also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or is to be output.

[0077] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A multi-factor wave forecasting method based on signal decomposition, characterized in that: The method comprises: Acquire ocean wave data, and decompose the ocean wave data into effective wave height time series data, effective wave period time series data, and wave direction time series data; Perform adaptive noise complete empirical mode decomposition on the effective wave height time series data, the effective wave period time series data, and the wave direction time series data, and calculate the entropy value of each intrinsic mode function obtained by decomposition; set the entropy threshold, and take the intrinsic mode function with an entropy value greater than the entropy threshold as the high-frequency mode function, and take the intrinsic mode function with an entropy value less than the entropy threshold as the low-frequency mode function; The low-frequency modal functions corresponding to the effective wave height time series data, effective wave period time series data, and wave direction time series data are divided into K 11 Group mode, K 21 Group mode, K 31 Group modality; The high-frequency modal functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data are superimposed and then subjected to variational modal decomposition. The modal functions obtained after variational modal decomposition are divided into K 12 Group mode, K 22 Group mode, K 32 Group modality; K 11 Group mode, K 12 The group modes are input into the BiLSTM model in sequence, and the effective wave height prediction results are output; 21 Group mode, K 22 The group modes are input into the LSTM model in sequence, and the prediction results of the effective wave period are output; K 31 Group mode, K 32 The group modes are input into the LSTM-GRU model in sequence, and the wave direction prediction results are output.

2. The multi-factor wave forecasting method based on signal decomposition according to claim 1, characterized in that: The process of performing adaptive noise complete empirical mode decomposition on the effective wave height time series data, the effective wave period time series data, and the wave direction time series data includes: The adaptive noise complete empirical mode decomposition is performed on the effective wave height time series data to obtain several first intrinsic mode functions; the adaptive noise complete empirical mode decomposition is performed on the effective wave period time series data to obtain several second intrinsic mode functions; the adaptive noise complete empirical mode decomposition is performed on the wave direction time series data to obtain several third intrinsic mode functions; Calculate the entropy value corresponding to each first eigenmode function, the second eigenmode function, and the third eigenmode function, and take the eigenmode function whose entropy value is less than the entropy threshold as the low-frequency mode function; The low-frequency modal function corresponding to the effective wave height time series data is divided into K 11 The low-frequency modal function corresponding to the effective wave period time series data is divided into K 21 Group mode, divide the low-frequency mode function corresponding to the wave direction time series data into K 31 Group modality.

3. A multi-factor ocean wave forecasting method based on signal decomposition according to claim 1 or 2, characterized in that: The process of calculating the entropy value of each eigenmode function includes: The intrinsic mode functions are processed by a low-pass Butterworth filter to obtain sub-time series of each intrinsic mode function at several scales; The sample entropy of each sub-time series of each intrinsic mode function at each scale is calculated, and the sum of the sample entropies at each scale is taken as the entropy value of the intrinsic mode function.

4. The multi-factor wave forecasting method based on signal decomposition according to claim 1, characterized in that: The process of performing variational mode decomposition after superimposing the high-frequency mode functions corresponding to the effective wave height time series data, the effective wave period time series data, and the wave direction time series data includes: The high-frequency modal functions corresponding to the effective wave height time series data are superimposed, and several first modal functions are obtained after variational modal decomposition; the first modal function is divided into K 12 Group modality; The high-frequency modal functions corresponding to the effective wave period time series data are superimposed, and several second modal functions are obtained after variational mode decomposition; the second modal function is divided into K 22 Group modality; The high-frequency modal functions corresponding to the wave direction time series data are superimposed, and several third modal functions are obtained after variational mode decomposition; the third modal function is divided into K 32 Group modality.

5. A multi-factor ocean wave forecasting method based on signal decomposition according to claim 2 or 4, characterized in that: The low-frequency modal functions corresponding to the effective wave height time series data, effective wave period time series data, and wave direction time series data are divided into K 11 Group mode, K 21 Group mode, K 31 Group mode; the mode functions obtained after variational mode decomposition are divided into K 12 Group mode, K 22 Group mode, K 32 The process of group modality includes: The low-frequency modal functions corresponding to the effective wave height time series data are divided, and the low-frequency modal functions with entropy values ​​greater than 1.40 are taken as intermediate-frequency IMFs, the low-frequency modal functions with entropy values ​​less than 1.40 and greater than 1.39 are taken as low-frequency IMFs, and the low-frequency modal functions with entropy values ​​less than 1.39 are taken as trend term IMFs. The first mode function corresponding to the effective wave height time series data is divided, and the first mode function with an entropy value greater than 9.1 is regarded as a high-frequency mode, the first mode function with an entropy value less than 9.1 and greater than 7.4 is regarded as a medium-frequency mode, and the first mode function with an entropy value less than 7.4 is regarded as a low-frequency mode; The low-frequency modal functions corresponding to the effective wave period time series data are divided, and the low-frequency modal functions with entropy values ​​greater than 1.20 are taken as intermediate-frequency IMFs, the low-frequency modal functions with entropy values ​​less than 1.20 and greater than 0.43 are taken as low-frequency IMFs, and the low-frequency modal functions with entropy values ​​less than 0.43 are taken as trend term IMFs; The second mode function corresponding to the effective wave period time series data is divided, and the second mode function with an entropy value greater than 11.9 is used as a high-frequency mode, the second mode function with an entropy value less than 11.9 and greater than 8.0 is used as a medium-frequency mode, and the second mode function with an entropy value less than 8.0 is used as a low-frequency mode; The low-frequency modal functions corresponding to the wave direction time series data are divided, and the low-frequency modal functions with entropy values ​​greater than 1.80 are taken as intermediate-frequency IMFs, the low-frequency modal functions with entropy values ​​less than 1.80 and greater than 1.67 are taken as low-frequency IMFs, and the low-frequency modal functions with entropy values ​​less than 1.67 are taken as trend term IMFs; The third mode function corresponding to the wave direction time series data is divided, and the third mode function with an entropy value greater than 9.2 is used as the high-frequency mode, the third mode function with an entropy value less than 9.2 and greater than 7.1 is used as the medium-frequency mode, and the third mode function with an entropy value less than 7.1 is used as the low-frequency mode.

6. The multi-factor wave forecasting method based on signal decomposition according to claim 1, characterized in that: The process of outputting the prediction results of effective wave height, effective wave period and wave direction includes: K 11 Group mode, K 12 The group modes are sequentially input into the pre-trained BiLSTM model, and the prediction sub-results corresponding to each group mode are added together as the significant wave height prediction result; K 21 Group mode, K 22 The group modes are input into the pre-trained LSTM model in sequence, and the prediction sub-results corresponding to each group mode are added together as the prediction result of the effective wave period; K 31 Group mode, K 32 The group modalities are input into the pre-trained LSTM-GRU model in sequence, and the prediction sub-results corresponding to each group modality are added together as the wave direction prediction result.

7. A multi-factor ocean wave forecasting method based on signal decomposition according to claim 1 or 6, characterized in that: The LSTM-GRU model includes an input layer, a first long short-term memory network, a first random inactivation layer, a first gated recurrent unit, a second random inactivation layer, a second long short-term memory network, a third random inactivation layer, and a fully connected layer connected in sequence.

8. An electronic device, comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the multi-factor wave forecasting method based on signal decomposition as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the multi-factor ocean wave forecasting method based on signal decomposition as described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the multi-factor ocean wave forecasting method based on signal decomposition as described in any one of claims 1 to 7 is implemented.

Images