A sea wave height prediction method and system

Abstract

The application discloses a sea wave height prediction method and system, and relates to the technical field of ocean prediction.The method comprises the following steps: a data acquisition step, a data preprocessing step, an EEMD decomposition step, a data division step, a network training step, a training end determination step and a wave height data prediction step.The application is helpful to improve the sea wave prediction accuracy without wasting a large amount of manpower cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of marine prediction, and in particular to a sea wave height prediction method and system. BACKGROUND

[0002] Significant wave height is the actual wave height value statistically in a certain way, which is an important parameter for measuring marine climate conditions and plays an important role in ship design, marine engineering construction and marine disaster prediction. How to accurately and quickly predict the significant wave height has become a problem in the development of modern marine technology. Numerical calculation prediction method relies on complex physical system and uses a large amount of observation data for prediction, which has problems of long calculation time and high prediction cost.

[0003] Machine learning method can learn the nonlinear characteristics of data and can quickly and accurately predict the significant wave height. However, machine learning method depends on feature engineering, and if good features cannot be learned, the prediction effect of the model will be significantly reduced.

[0004] In recent years, the development of computer technology makes it possible to apply complex models. Deep learning method can automatically learn data features and is widely used in marine prediction field and has achieved good results, such as SWAN-LSTM, CNN-LSTM, ConvLSTM, etc.

[0005] However, the above methods do not fully mine the signal characteristics of the significant wave height. Marine data has nonlinear, asymmetric and non-stationary characteristics, in order to reduce the non-stationary characteristics of marine data, the original data sequence is decomposed into different scale modal components by using signal decomposition technology, and then each modal component is modeled and predicted.

[0006] EOFWNN algorithm is proposed by combining empirical orthogonal function, wavelet analysis and neural network, and the wave height of eight wave observation stations along the coast of Japan is predicted. Compared with WNN model, EOWNN model performs better in all lead times regardless of the decomposition level of wavelet analysis. The frequency signal obtained by applying EMD algorithm to the decomposition of significant wave height is applied to the LSTM network prediction, and good results are obtained. Therefore, under the condition of ensuring the accuracy of decomposition, it is still a problem to be solved to further improve the prediction accuracy and reduce the calculation cost.

[0007] Therefore, a sea wave height prediction method and system are proposed to solve the problems existing in the prior art, which is a problem urgently to be solved by those skilled in the art. SUMMARY

[0008] Therefore, a sea wave height prediction method and system are proposed to solve the problems existing in the prior art, which is a problem urgently to be solved by those skilled in the art.

[0009] In order to achieve the above object, the present application adopts the following technical solutions:

[0010] The sea wave height prediction method based on EEMD and Seq-to-Seq model comprises the following steps:

[0011] The data acquisition step: selecting a wave height data set;

[0012] The data preprocessing step: after the missing data in the wave height data set is completed, normalization processing is performed to obtain a preprocessed wave height data set;

[0013] The EEMD decomposition step: the preprocessed wave height data set is decomposed by EEMD to obtain decomposed basic modal components IMF;

[0014] The data division step: the decomposed basic modal components IMF are divided into training data sets and test data sets according to a certain proportion, and the training data sets and test data sets are obtained;

[0015] The network training step: the data of the training data set is used to train the network, the loss function used in the training is the mean square error MSE between the prediction result and the true value, the back propagation is performed through the gradient descent method, and the trainable parameters in the network are updated;

[0016] The training end determination step: whether the trained network meets the parameter requirement is determined, if not, the training data set selection step is returned; if yes, the network training step is ended, and the trained Seq-to-Seq model based on LSTM and attention mechanism is outputted;

[0017] The wave height data prediction step: the test data set is inputted into the trained Seq-to-Seq model based on LSTM and attention mechanism, each decomposed basic modal component IMF is obtained, the prediction value component is obtained, and finally the prediction results on all decomposed basic modal components IMF are summed to obtain the predicted sea wave height value.

[0018] The above method, optionally, the formula of the normalization processing in the data preprocessing step is as follows:

[0019]

[0020] Wherein, X N is the normalized result, and X represents the input data.

[0021] The above method, optionally, in the EEMD decomposition step, the EEMD decomposition step is specifically as follows:

[0022] S101 inputting wave height data and initializing the number of trials M;

[0023] S102 in mth In the second test, a white noise sequence n m (t) is added to the original data x(t), and the formula is as follows:

[0024] x m = x(t) + n m (t) (2)

[0025] S103 performs EMD decomposition on the newly obtained data x m (t) to obtain the basic modal component IMF j denoted as C j,m , and C j,m is the jth IMF in the m th th test.

[0026] S104 determines whether it is the Mth test, and if not, repeats S102-S103, and adds different white noise each time.

[0027] The IMF averages of all tests are obtained, as shown in the formula:

[0028]

[0029] wherein IMF j represents the jth IMF average of EMD decomposition.

[0030] In the above method, optionally, the model in the network training step is a Seq-to-Seq model structure based on the combination of LSTM and attention mechanism, the LSTM encoder with j time steps outputs the result of each time step, the attention value is calculated with the (i-1)th time step output by the LSTM decoder, and the result is obtained as the input of the i th time step of the decoder LSTM, i represents the i th time of prediction.

[0031] In the above method, optionally, in the training end determination step, the LSTM layers used by the encoder and the decoder each use 100 neurons, and a fully connected layer with 1 neuron after the decoder is used as the output, and the output value ranges from [0, +∞];

[0032] The loss function used by the model is mean square error MSE, and the Adam optimizer is used, and the learning rate is set to 0.001.

[0033] In the above method, optionally, in the training end determination step, the formula of MSE is as follows:

[0034]

[0035] wherein y pre is the predicted value, and y trueThe real value is n, and the total number of prediction samples is n.

[0036] The method described above, optionally, after the training end determination step, further includes a model performance evaluation step, and the specific content is:

[0037] Using LSTM and attention mechanism, taking 3 to 72 hours in the future as the benchmark, and taking RMSE, MAE, MAPE, and Pearson as evaluation indexes to verify the performance of the Seq-to-Seq model based on LSTM and attention mechanism.

[0038] The method described above, optionally, in the data division step, the first 75% of the decomposed basic modal component IMF is the training data set, and the last 25% of the decomposed basic modal component IMF is the test data set.

[0039] The sea wave height prediction system based on EEMD and Seq-to-Seq model applies the sea wave height prediction method based on EEMD and Seq-to-Seq model described above, including a data acquisition module, a data preprocessing module, an EEMD module, a data division module, a network training module, a Seq-to-Seq prediction module based on LSTM and attention mechanism, and a wave height data prediction value output module.

[0040] The data acquisition module is used to select the wave height data set.

[0041] The data preprocessing module is connected with the output end of the data acquisition module, and is used to complete the missing data in the wave height data set and then perform normalization processing to obtain the preprocessed wave height data set.

[0042] The EEMD module is connected with the output end of the data preprocessing module, and is used to perform EEMD decomposition on the preprocessed wave height data set to obtain the decomposed basic modal component IMF.

[0043] The data division module is connected with the output end of the EEMD module, and is used to divide the decomposed basic modal component IMF into a training data set and a test data set according to a certain proportion to obtain the training data set and the test data set.

[0044] The network training module is connected with the first output end of the data division module, and is configured to train the network by using the data of the training data set. The loss function used in the training is the mean square error (MSE) between the prediction result and the true value. The gradient descent method is used for back propagation to update the trainable parameters in the network. It is determined whether the trained network meets the parameter requirement. If not, the training data set selection step is returned. If yes, the network training step is ended, and the trained Seq-to-Seq model based on the LSTM and the attention mechanism is output. The trained Seq-to-Seq model based on the LSTM and the attention mechanism is output.

[0045] The Seq-to-Seq prediction module based on the LSTM and the attention mechanism is connected with the second output end of the data division module, and is configured to input the test data set into the trained Seq-to-Seq model based on the LSTM and the attention mechanism, obtain the prediction value component for each decomposed IMF, and sum the prediction results on all decomposed IMFs to obtain the predicted wave height value.

[0046] The wave height data prediction value output module is connected with the output end of the Seq-to-Seq prediction module based on the LSTM and the attention mechanism, and is configured to output the obtained predicted wave height value.

[0047] The system described above can further include a model performance evaluation module connected with the output end of the Seq-to-Seq prediction module based on the LSTM and the attention mechanism, and configured to verify the performance of the Seq-to-Seq model based on the LSTM and the attention mechanism.

[0048] A storage medium includes stored instructions, wherein the instructions, when executed, control a device in which the storage medium is located to perform the wave height prediction method based on the EEMD and the Seq-to-Seq model.

[0049] An electronic device includes a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to be executed by one or more processors to perform the wave height prediction method based on the EEMD and the Seq-to-Seq model.

[0050] Compared with the prior art, the sea wave height prediction method and system based on EEMD and Seq-to-Seq model provided by the application have the following advantages: 1) the model is superior to other models in that the original data is decomposed by EEMD, a non-stationary wave height sequence is converted into a wave height component with a clear mode, the component has predictability, the prediction accuracy can be significantly improved, and the mode aliasing problem caused by EMD is solved; 2) at the same time, the Seq-to-Seq structure can reduce prediction bias and error by considering the output sequence dependency, while the traditional LSTM model does not have this feature, which greatly improves the sea wave height prediction accuracy; 3) the proposed hybrid model based on EEMD and Seq-to-Seq is suitable for prediction of other marine element sequence data. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of the provided drawings.

[0052] Figure 1 The sea wave height prediction method based on EEMD and Seq-to-Seq model provided by the present application has the flow chart;

[0053] Figure 2 The EMD decomposition flow chart provided by the present application has the flow chart;

[0054] Figure 3 The EEMD decomposition flow chart provided by the present application has the flow chart;

[0055] Figure 4 The LSTM layer structure diagram provided by the present application has the flow chart;

[0056] Figure 5 The prediction model block diagram provided by the present application has the flow chart, wherein 5.1 is the Seq-to-Seq model block diagram based on LSTM and attention mechanism, and 5.2 is the attention mechanism model block diagram;

[0057] Figure 6 The sea wave height prediction system structure block diagram based on EEMD and Seq-to-Seq model provided by the present application has the flow chart;

[0058] Figure 7 The experimental result diagram provided by the embodiment of the present application has the flow chart;

[0059] Figure 8 The electronic device provided by the embodiment of the present application has the flow chart. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0061] In the present application, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0062] The present application can be used in a plurality of general or special-purpose computing device environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor devices, distributed computing environments that include any of the above devices or devices, and the like.

[0063] Reference Figure 1 As shown in the figure, the present application discloses a sea wave height prediction method based on EEMD and Seq-to-Seq model, comprising the following steps:

[0064] Data acquisition step: select wave height data set;

[0065] Data preprocessing step: after the missing data in the wave height data set is completed, normalization processing is performed to obtain the preprocessed wave height data set;

[0066] EEMD decomposition step: the preprocessed wave height data set is decomposed by EEMD to obtain the decomposed intrinsic mode component IMF;

[0067] Data division step: the decomposed intrinsic mode component IMF is divided into training data set and test data set according to a certain proportion, to obtain the training data set and test data set;

[0068] Network training step: the network is trained using the data of the training data set, the loss function used in the training is the mean square error (MSE) between the predicted result and the true value, and the trainable parameters in the network are updated through back propagation by gradient descent method;

[0069] Training end determination step: determine whether the trained network meets the parameter requirements, if not, return to the training data set selection step; if yes, end the network training step and output the trained Seq-to-Seq model based on LSTM and attention mechanism;

[0070] Wave height data prediction step: input the test data set into the trained Seq-to-Seq model based on LSTM and attention mechanism, get the predicted value component for each decomposed basic modal component IMF, and finally sum the predicted results on all decomposed basic modal components IMF to get the predicted sea wave height value.

[0071] Further, the formula of the normalization processing in the data preprocessing step is as follows:

[0072]

[0073] Where, X N is the normalized result, and X represents the input data.

[0074] Further, as shown in Figure 3 , in the EEMD decomposition step, the EEMD decomposition step is as follows:

[0075] S101 input wave height data and initialize the number of trials M;

[0076] S102 in the m th th trial, add a white noise sequence n m (t) to the original data x(t), the formula is:

[0077] x m = x(t) + n m (t) (2)

[0078] S103 EMD decomposes the newly obtained data x m (t) to obtain the basic modal component IMF j , denoted as C j,m , C j,m is the jth IMF in the m th th trial;

[0079] S104 determine whether it is the Mth trial, if not, repeat S102-S103, and add different white noise each time;

[0080] The IMF mean is obtained by averaging all the tested IMF, as shown in the formula:

[0081]

[0082] Where, IMF j is the jth IMF mean of EMD decomposition.

[0083] Specifically, referring to Figure 2 , the steps of the EMD algorithm are as follows:

[0084] S201 finds the local extreme value of the original data x(t) and uses the local extreme value to do spline interpolation to obtain the upper envelope e max (t) and the lower envelope e min (t);

[0085] S202 takes the average value m(t) of the upper and lower envelopes, as shown in the formula:

[0086]

[0087] S203 subtracts the average value m(t) from the original function x(t) to obtain a new function h(t), as shown in the formula:

[0088] h(t) = x(t) - m(t) (6)

[0089] S204 judges whether h(t) satisfies the stopping criterion (0.2 < standard deviation < 0.3) of the screening algorithm. If the stopping criterion is satisfied, h(t) is the IMF and the next step is executed. If it is not satisfied, h(t) is taken as the original input sequence to repeat steps 1 and 2 until the condition is satisfied. Define h(t) that satisfies the condition as IMF, denoted as C1(t). Subtract the extracted IMF from the original data to calculate as shown in the formula:

[0090] r1(t) = x(t) - C1(t) (7)

[0091] S205 replaces the original function x(t) with the new function r1(t) to repeat the above steps until r n (t) cannot be decomposed. After EMD decomposition, the original signal x(t) is decomposed into multiple IMFs and one residual, represented as:

[0092]

[0093] Where n is the number of basic modal components IMF, r n is the residual of the data x(t), and IMF is a simple oscillation function with different amplitudes and frequencies.

[0094] Further, in the data division step, the first 75% of the decomposed intrinsic mode components IMF is the training data set, and the last 25% of the decomposed intrinsic mode components IMF is the test data set; the divided data is arranged into input data of 6 consecutive groups and output data of corresponding time steps. The shape of the arranged data should be:

[0095] Input: [batchsize, 6, 1]; Output: [batchsize, predicted time steps, 1].

[0096] Since the input sequence length is 6, the encoder consists of 6 LSTM cells, and the LSTM cells of the decoder are determined according to the prediction length. The specific steps are as follows:

[0097] The preprocessed data is input into the LSTM according to the batchsize, and the dimension of the LSTM is set to 100. The first step in the LSTM is to determine whether some information of the previous time step should be lost. This decision is made through the forget gate. The gate reads Sh t-1 and x t , and outputs a value between 0 and 1 to each number in the cell state SC t-1 after passing through a sigmoid function. 1 represents “complete retention”, and 0 represents “complete forgetting” as shown in the formula below: Figure 4

[0098] f t = σ(w f *[Sh t-1 ,x t ]+b f ) (9)

[0099] Where f t represents the output of the forget gate, w f and b f represent the weight and bias, Sh t-1 represents the output of the previous time stamp, x t represents the input of the current time step, and σ represents the sigmoid activation function.

[0100] The next step is to determine which new information will be stored in the cell state. It contains two parts: the first sigmoid activation function is called the “input gate layer” to determine what values to update. The second tanh activation function creates a new candidate value vector which is used to update the state. The formula is as follows:

[0101] i t = σ(W i *[Sh t-1 ,x​t ]+b i ) (10)

[0102]

[0103] where i t denotes the output of the input gate, denotes the candidate vector, W i , b i are the weight and bias of the input gate, W c is the weight of the candidate vector, x t is the input of the current time step.

[0104] The SC t-1 is updated to SC t , which is as follows:

[0105]

[0106] where SC t is the state vector of the current time step.

[0107] The final decision output, the output is first based on the current cell state. First run a sigmoid to determine part of the information of the output, then the cell state is processed by tanh and multiplied by the output of the output gate to obtain the final output information.

[0108] o t = σ(W o *[Sh t-1 ,x t ]+b o ) (13)

[0109] Sh t = o t *tanh(SC t ) (14)

[0110] where o t is the output of the output gate, W o , b o are the weight and bias of the output gate, Sh t is the output of the current time step.

[0111] Further, the model in the network training step is a Seq-to-Seq model structure based on the combination of LSTM and attention mechanism, and the LSTM encoder with j time steps outputs the results of each time step. The attention value is calculated with the i-1 time step output by the LSTM decoder, and the result is obtained as the input of the i time step decoder LSTM, i represents the i th moment of prediction.

[0112] Specifically, based on a Seq-to-Seq model structure combining LSTM and Attention, the output of the LSTM encoder at each of j time steps is used to calculate the attention value with the output of the LSTM decoder at the (i-1)th time step. The result is then used as the input to the LSTM decoder at the ith time step. Since the lengths of the input and output sequences may differ, this Seq-to-Seq structure addresses the issue of different time steps. LSTM layers are available as standard packages; therefore, this project utilizes the built-in LSTM layer component within the Keras framework. The specific formula is as follows:

[0113] f t =σ(w f *[Sh t-1 ,x t ]+b f )

[0114] i t =σ(W i *[Sh t-1 ,x t ]+b i )

[0115]

[0116]

[0117] o t =σ(W o *[Sh t-1 ,x t ]+b o )

[0118] Sh t =O t *tanh(SC t )

[0119]

[0120] c i =∑ j α ij *Sh j

[0121] Among them, f t Indicates the output of the forget gate, i t Indicates the output of the input gate. For candidate vectors, o t For the output of the output gate, SC t The state vector for each time step, Sh t X is the output of the current time step.t is the input, c i is the attention value, a ij is the attention weight.

[0122] Specifically, the dimension of the LSTM output is defined as 100, and if the return_sequences parameter is set to True, the Sh t , if the return_state parameter is set to True, the Sh t and SC t , and then the attention mechanism is calculated.

[0123] First, the similarity function F is used to calculate the similarity between the jth data of the input sequence and the ith data of the output sequence (common similarity calculations include dot product, cosine, concatenation, and multi-layer perception (MLP)). Here, MLP is selected, and then the weight is calculated by the softmax function. Finally, the weight and the encoder hidden state vector Sh i are weighted and summed to obtain the attention value C i , and the specific formula is as follows:

[0124]

[0125] C i =∑ j α ij Sh j (16)

[0126] where a ij is the attention weight, C i is the attention value, F is the similarity calculation, Sh j is the encoder output vector, and SH i-1 is the decoder output vector.

[0127] The obtained attention value and the SC i of the previous timestamp are sent to the LSTMCell of the decoder together with Sh i for operation as shown in Figure 5 , where 5.1 is a Seq-to-Seq model block diagram based on LSTM and attention mechanism, and 5.2 is an attention mechanism model block diagram.

[0128] Furthermore, in the training end determination step, the LSTM layers used by the encoder and the decoder each use 100 neurons, and a fully connected layer with one neuron and an output value range of [0, +∞] is used after the decoder as the output.

[0129] The loss function used in the model is mean squared error (MSE), and the Adam optimizer is used with a learning rate of 0.001.

[0130] Furthermore, the formula for MSE in the training completion determination step is as follows:

[0131]

[0132] Among them, y pre For the predicted value, y true is the true value, and n is the total number of predicted samples.

[0133] Furthermore, after the training completion determination step, a model performance evaluation step is also included, the specific content of which is as follows:

[0134] Using LSTM and attention mechanisms, the performance of the LSTM-based Seq-to-Seq model was validated based on predictions over the next 3 to 72 hours, with RMSE, MAE, MAPE, and Pearson as evaluation metrics.

[0135] Specifically, the formulas for RMSE, MAE, MAPE, and Pearson's equations are as follows:

[0136]

[0137]

[0138]

[0139]

[0140] Where, x i For the predicted value, y i The true value is n, where n is the number of samples. The average of the predicted values. This is the average of the true values.

[0141] and Figure 1 Corresponding to the method described above, this embodiment of the invention also provides a wave height prediction system based on EEMD and Seq-to-Seq models, used for... Figure 1 The specific implementation of the method, the wave height prediction system based on EEMD and Seq-to-Seq models provided in this embodiment of the invention, can be applied to computer terminals or various mobile devices. Its structural diagram is shown below. Figure 6 As shown, it specifically includes: a data acquisition module, a data preprocessing module, an EEMD module, a data partitioning module, a network training module, a Seq-to-Seq prediction module based on LSTM and attention mechanism, and a wave height data prediction output module;

[0142] a data acquisition module configured to select a wave height data set;

[0143] a data preprocessing module connected to an output end of the data acquisition module, configured to complete missing data in the wave height data set and perform normalization processing to obtain a preprocessed wave height data set;

[0144] an EEMD module connected to an output end of the data preprocessing module, configured to perform EEMD decomposition on the preprocessed wave height data set to obtain decomposed intrinsic mode components IMF;

[0145] a data division module connected to an output end of the EEMD module, configured to divide the decomposed intrinsic mode components IMF into a training data set and a test data set according to a certain proportion;

[0146] a network training module connected to a first output end of the data division module, configured to train the network using data of the training data set, wherein a loss function used in the training is a mean square error MSE between a prediction result and an actual value, and the network is updated by back propagation through a gradient descent method; whether the trained network meets parameter requirements is determined, if not, the step of selecting the training data set is returned; if yes, the step of training the network is ended, and a trained Seq-to-Seq model based on LSTM and an attention mechanism is outputted; and the trained Seq-to-Seq model based on LSTM and the attention mechanism is outputted;

[0147] a Seq-to-Seq prediction module based on LSTM and the attention mechanism, connected to a second output end of the data division module, configured to input the test data set and each decomposed intrinsic mode component IMF into the trained Seq-to-Seq model based on LSTM and the attention mechanism to obtain a prediction value component, and finally sum prediction results on all decomposed intrinsic mode components IMF to obtain a predicted sea wave height value;

[0148] a wave height data prediction value output module connected to an output end of the Seq-to-Seq prediction module based on LSTM and the attention mechanism, configured to output the obtained predicted sea wave height value.

[0149] Further, a model performance evaluation module is further included, connected to an output end of the Seq-to-Seq prediction module based on LSTM and the attention mechanism, configured to verify performance of the Seq-to-Seq model based on LSTM and the attention mechanism.

[0150] Further, the data division module is configured to divide the decomposed intrinsic mode components IMF into a training data set with the first 75% of the decomposed intrinsic mode components IMF and a test data set with the last 25% of the decomposed intrinsic mode components IMF.

[0151] In one specific embodiment, the method disclosed herein is simulated.

[0152] The data source is the 41040 buoy of NDBC (located in the Atlantic Ocean surface east of the Bahamas), which obtains the wave height data of the sea surface in units of 1 hour, and the data may be lost due to various forces of nature during the acquisition process, and in this experiment, the lost data is completed by quadratic curve interpolation. The completed data is normalized. 75% of the normalized data is used as the training set, and 25% is used as the test set. The hyperparameters used in training are shown in Table 1.

[0153] Table 1

[0154]

[0155] And in the experiment, several commonly used prediction algorithms are compared, and the results are shown in Table 2. The results show that the model proposed in the present application has reached the optimal accuracy at each step, which shows that the prediction modeling of the present application for sea wave height is very effective.

[0156] Table 2

[0157]

[0158]

[0159] In order to more objectively compare the effectiveness of different algorithms, Figure 7 The scatter plot of the predicted value and the true value of different algorithms at the prediction window of 3, 6, 12 and 24h is shown. It can be seen that the hybrid model based on EEMD and Seq-to-Seq proposed in the present application is more accurate in fitting the true value.

[0160] The embodiment of the present application also provides a storage medium, which comprises stored instructions, wherein when the instructions are executed, the device where the storage medium is located performs the sea wave height prediction method based on the EEMD and Seq-to-Seq model.

[0161] The embodiment of the present application also provides an electronic device, and a structure diagram thereof is shown as Figure 8 Specifically, the electronic device comprises a memory 801 and one or more instructions 802, wherein the one or more instructions 802 are stored in the memory 801 and are configured to be executed by one or more processors 803 to perform the sea wave height prediction operation based on the EEMD and Seq-to-Seq model.

[0162] The various embodiments described in this specification are described in progressive order of complexity, from the simplest embodiment to more complex embodiments. Embodiments described in this specification can be understood, and implemented, by those skilled in the art with the benefit of the following description and practice of the disclosure without necessitating use of more complex embodiments. The same parts and features of the various embodiments described in this specification can be referred to by the same reference numerals in all figures of which their presence in some or all of the various embodiments is indicated, but not necessarily by the same reference designators in all figures. Each embodiment described in this specification highlights differences from other embodiments, and each embodiment described in this specification highlights differences from other embodiments. In particular, for system or system embodiments, since they are substantially similar to method embodiments, they are described more simply, and reference is made to the description of the method embodiments. The systems and system embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e. they may be located in one place, or distributed over multiple network units. Some or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0163] The skilled person will further appreciate that the elements and algorithm steps of the examples described in relation to the embodiments disclosed herein can be implemented in electronic hardware, computer software or a combination of both.

[0164] To clearly illustrate the interchangeability of hardware and software, the elements and steps of the examples described above have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0165] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A sea wave height prediction method based on EEMD and Seq-to-Seq model, characterized in that, The method comprises the following steps: a data acquisition step of selecting a wave height data set; a data preprocessing step of completing missing data in the wave height data set and then performing normalization processing to obtain a preprocessed wave height data set; an EEMD decomposition step of decomposing the preprocessed wave height data set to obtain decomposed intrinsic mode components IMF; a data division step of dividing the decomposed intrinsic mode components IMF into a training data set and a test data set according to a certain proportion; a network training step of training the network using the data of the training data set, wherein a loss function used in the training is a mean square error MSE between a prediction result and an actual value, and the network is updated by back propagation through a gradient descent method; a training end determination step of determining whether the trained network meets parameter requirements, and if not, returning to the training data set selection step, and if so, ending the network training step and outputting a trained Seq-to-Seq model based on LSTM and an attention mechanism; a wave height data prediction step of inputting the test data set into the trained Seq-to-Seq model based on LSTM and the attention mechanism, obtaining a prediction component for each decomposed intrinsic mode component IMF, and summing prediction results on all decomposed intrinsic mode components IMF to obtain a predicted sea wave height value; The model in the network training step is a Seq-to-Seq model structure based on the combination of LSTM and attention mechanism, which has j The LSTM encoder outputs the result of each time step, and the attention value is calculated with the output of the first i- 1 time step of the LSTM decoder, and the result is taken as the input of the decoder LSTM of the first i time step, i which represents the predicted first i moment.

2. The sea wave height prediction method based on EEMD and a Seq-to-Seq model according to claim 1, wherein a normalization processing formula in the data preprocessing step is as follows:

3. The sea wave height prediction method based on EEMD and a Seq-to-Seq model according to claim 2, wherein the EEMD decomposition step is specifically as follows: （1） wherein, is the normalized result, represents the input data. The IMF mean value is obtained by averaging all test IMFs, as shown in the formula:

4. The sea wave height prediction method based on EEMD and a Seq-to-Seq model according to claim 1, wherein in the training end determination step, the LSTM layer used in the encoder and the decoder uses 100 neurons, and a fully connected layer with one neuron and a number of 1 is used as an output after the decoder, and the output value ranges from 0 to +∞; S101 input wave height data, and initialize the number of trials ; S102 In the secondary experiment, white noise sequences are added to the original data , and the formula is: （2） S103 To newly obtained data EMD decomposition, the basic modal component obtained denoted as , is The jth IMF in the nth test; S104 judges whether it is the first time If not, repeat S102-S103, and add different white noise each time. The loss function used in the model is a mean square error MSE, and an Adam optimizer is used with a learning rate of 0.

001. （3） wherein, denotes the mean value of the jth IMF of the EMD decomposition.

5. The sea wave height prediction method based on EEMD and a Seq-to-Seq model according to claim 4, wherein a formula of the MSE in the training end determination step is as follows:

6. The sea wave height prediction method based on EEMD and a Seq-to-Seq model according to claim 1, wherein the method further comprises a model performance evaluation step after the training end determination step, and the specific content is as follows: ​ ​ ​ （4） wherein, is the predicted value, is the true value, n is the total number of prediction samples. ​ ​ The LSTM and attention mechanism are used to predict the future 3 to 72 hours as the benchmark, and the performance of the Seq-to-Seq model based on the LSTM and attention mechanism is verified by taking RMSE, MAE, MAPE, and Pearson as evaluation indexes.

7. The sea wave height prediction method based on EEMD and Seq-to-Seq model according to claim 1, characterized in that, In the data division step, the first 75% of the decomposed intrinsic mode components IMF is used as the training data set, and the last 25% of the decomposed intrinsic mode components IMF is used as the test data set.

8. A sea wave height prediction system based on EEMD and Seq-to-Seq model, characterized in that The sea wave height prediction method based on EEMD and Seq-to-Seq model according to any one of claims 1-7 comprises a data acquisition module, a data preprocessing module, an EEMD module, a data division module, a network training module, a Seq-to-Seq prediction module based on LSTM and attention mechanism, and a wave height data prediction value output module. The data acquisition module is used to select the wave height data set. The data preprocessing module is connected to the output end of the data acquisition module and is used to complete the missing data in the wave height data set and then perform normalization processing to obtain the preprocessed wave height data set. The EEMD module is connected to the output end of the data preprocessing module and is used to perform EEMD decomposition on the preprocessed wave height data set to obtain the decomposed intrinsic mode components IMF. The data division module is connected to the output end of the EEMD module and is used to divide the decomposed intrinsic mode components IMF into the training data set and the test data set according to a certain proportion to obtain the training data set and the test data set. The network training module is connected to the first output end of the data division module and is used to train the network by using the data of the training data set. The loss function used in the training is the mean square error MSE between the prediction result and the true value. The trainable parameters in the network are updated by using the gradient descent method for back propagation. It is determined whether the trained network meets the parameter requirement. If not, the training data set selection step is returned. If yes, the network training step is ended, and the trained Seq-to-Seq model based on LSTM and attention mechanism is outputted. The Seq-to-Seq prediction module based on LSTM and attention mechanism is connected to the second output end of the data division module and is used to input the test data set into the trained Seq-to-Seq model based on LSTM and attention mechanism to obtain the prediction value component for each decomposed intrinsic mode component IMF. Finally, the prediction results on all decomposed intrinsic mode components IMF are summed to obtain the predicted sea wave height value. The wave height data prediction value output module is connected to the output end of the Seq-to-Seq prediction module based on LSTM and attention mechanism and is used to output the obtained predicted sea wave height value.

9. The sea wave height prediction system based on EEMD and Seq-to-Seq model according to claim 8, characterized in that, The model performance evaluation module is further connected with the output end of the Seq-to-Seq prediction module based on the LSTM and the attention mechanism, and is used for verifying the performance of the Seq-to-Seq model based on the LSTM and the attention mechanism.

Images