CN117371303A - Prediction method for effective wave height under sea wave - Google Patents

Abstract

The invention provides a method for predicting effective wave height under sea waves, which includes step 1: forming a data set based on historical NOAA buoy data; preprocessing the data set to filter out missing values and outlier data, and then dividing it into a training set and test set; Step 2: Use the Tensorflow architecture, call the LSTM layer from the keras library to build the LSTM model and then add the Attention layer to obtain a prediction model that combines the LSTM neural network and the Attention mechanism; Step 3: Use the training set to The forecast model is iteratively trained, and the test set is input into the trained forecast model for testing; and the accuracy of the forecast model is verified according to the denormalization process, and the results that meet the preset forecast accuracy standard evaluation are The forecast model serves as a significant wave height prediction model for sea waves and is used to determine the significant wave height of a point to be measured. This invention is used in the field of marine hydrometeorological forecasting, improves the accuracy of the LSTM model in predicting significant wave height, and lays the foundation for research on intelligent wave forecasting systems. .

Description

A method for predicting effective wave height under ocean waves

Technical field

The invention relates to the field of marine hydrometeorological forecasting, specifically a method for predicting effective wave height under sea waves.

Background technique

As one of the most important ocean phenomena, the study of waves is of vital significance to ensuring navigation safety, coastal activities and climate systems. The complex and random nature of waves brings great challenges to coastal and ocean research work. . Coastal and offshore engineers often use different methods such as field measurements, theoretical studies and numerical simulations to identify wave climate and extreme wave characteristics as well as annual properties of waves. In the fields of navigation and fishery, wave forecast plays an important role in resisting harsh sea conditions and ensuring operational safety. Waves are mainly described by factors such as wave height, wave period and wave direction, among which wave height occupies the primary position among wave parameters. At present, most research on wave height prediction is based on numerical simulation, but numerical model prediction has problems such as long calculation time, wide range, and high accuracy requirements.

The currently commonly used numerical prediction model is a numerical approximation model driven by physical laws, which implements wave prediction by solving physical equations through iterative calculations. The main forecasting tool of the intelligent forecasting system is the deep learning forecasting model, which is an intelligent forecasting model driven by big data. It uses deep learning methods to learn the spatiotemporal evolution rules of waves from historical wind and wave data to achieve wave forecasting.

Normally, during wave propagation, the effective wave height at each location is affected by some other features in addition to the effect of wind. However, it is necessary to consider how to select features. Therefore, in order to more accurately understand and master the method of intelligently forecasting effective wave heights based on deep learning, it is necessary to consider the impact of feature weights on forecasting and improve the accuracy of forecasting.

Contents of the invention

The purpose of the present invention is to provide a method for predicting effective wave height under ocean waves, which is used to solve the shortcomings of existing numerical models such as large amount of calculation, high cost, inability to predict quickly, and reliance on feature engineering, and to achieve low-cost, fast and accurate predict.

In order to achieve the above objects, the present invention provides the following technical solutions:

A method for predicting effective wave height under sea waves, including the following steps:

Step 1: Compose a data set based on historical NOAA buoy data; preprocess the data set to filter out missing values and outlier data, and divide it into a training set and a test set;

Step 2: Use the Tensorflow architecture, call the LSTM layer from the keras library to build the LSTM model and then add the Attention layer to obtain a forecast model that combines the LSTM neural network and the Attention mechanism;

Step 3: Use the training set to iteratively train the forecast model, and input the test set into the trained forecast model for testing; and perform the accuracy of the forecast model according to the denormalization process. Verify that the forecast model that meets the preset forecast accuracy standard evaluation will be used as the effective wave height forecast model of sea waves;

Step 4: After inputting the buoy data of the wave point to be measured into the output of the wave significant wave height prediction model, determine the significant wave height of the wave point to be measured.

Further, the data set in step one specifically includes wind direction, average wind speed, peak wind speed, dominant wave period, average wave period, wave direction, sea level pressure, air temperature, sea surface temperature and significant wave height.

Further, the input data of the training set and the test set in step 3 respectively include wind direction, average wind speed, peak wind speed, dominant wave period, average wave period, wave direction, sea level pressure, air temperature and sea surface temperature; The output data of the training set and test set respectively include significant wave heights.

Further, the specific step of dividing the data set in step one is to divide the data set into a training set and a test set at a ratio of 3:1 based on time series and year data.

Further, the specific training method of the ocean wave significant wave height prediction model includes:

Input the input data of the training set into the LSTM neural network, predict according to the data change pattern at the previous moment, and output the prediction information;

The LSTM prediction information is input to the added Attention layer. The Attention layer assigns probability weights to the output information of the LSTM neural network and dynamically adjusts the learning rate to improve the generalization of the effective wave height prediction model.

Further, the mathematical model of the LSTM neural network is:

I _t =σ(X _t W _xi +H _t-1 W _hi +b _i )

F _t =σ(X _t W _xf +H _t-1 W _hf +b _f );

O _t =σ(X _t W _xo +H _t-1 W _ho +b _o );

In the formula: I _t , F _t , O _t represent the input gate, forgetting gate and output gate respectively; X _t represents the input data, W _ij represents the weight matrix of the output gate, and H _t-1 represents the hidden state of the previous time step. , b _i represents the bias term of the output gate, σ represents the activation function, C _t represents candidate memory cells and memory cells respectively, where/> Represents the multiplication of matrix elements.

Further, the mathematical model of the Attention mechanism is:

e _t =σ(W _e x _t + _be )

In the formula: e _t , We _e , b _e , σ respectively represent the weight coefficient combination, trainable weight matrix, bias vector and activation function corresponding to each input data at the current moment;

Each attention weight coefficient is normalized through the sofimax function to obtain the attention weight, where α _{m, t} is the attention weight value of the m-th feature, expressed as:

Recompute the input feature vector x _t as weighted Vector, expressed as:

Further, the normalization processing calculation formula is:

Among them, X ^* is the standardized data, X is the original data, is the mean of the original data, and δ is the standard deviation of the original data. The processed feature data conforms to the standard normal distribution with a mean of 0 and a standard deviation of 1.

Further, the method for verifying the accuracy of the forecast model specifically includes:

Input the test set into the trained forecast model, and obtain the predicted effective wave height through denormalization processing;

Compare the predicted wave height with the effective wave height in the test set, and calculate the root mean square error, mean absolute error, mean absolute percentage error and goodness of fit as evaluation indicators to verify the accuracy of the model; the calculation formula is:

in, represents the deep learning forecast result; y represents the significant wave height in the NOAA data set;/> Represents the average significant wave height in the NOAA data set; N represents the number of observations; RMSE represents the root mean square error; MAE represents the mean absolute error; MAPE represents the absolute percentage error; R ² represents the goodness of fit.

Compared with the existing technology, the present invention adopts the above technical solution and has the following technical effects:

The invention provides a method for predicting effective wave height under ocean waves. First, a data set is formed based on historical NOAA buoy data; the data set is preprocessed to filter out missing values and outlier data, and then divided into a training set and a test set; and Using the Tensorflow architecture, call the LSTM layer from the keras library to build the LSTM model and then add the Attention layer to obtain a forecast model that combines the LSTM neural network and the Attention mechanism; input the training set to iteratively train the forecast model, and test the Sets are input into the trained forecast model for testing; and the accuracy of the forecast model is verified according to the denormalization process, and the forecast model that meets the preset forecast accuracy standard evaluation is used as the effective wave height forecast model; Using this prediction method of significant wave height can obtain:

(1) This invention is suitable for the prediction of sea wave significant wave height data. The designed prediction model adopts a network structure that combines the LSTM neural network and the Attention mechanism. Its advantages include: LSTM, which fully extracts the characteristics of time series data information and improves model prediction. Accuracy, suitable for prediction of time series data; the Attention mechanism can weight the prediction results of LSTM and extract the important parts of the prediction results.

(2) It solves the various shortcomings of existing numerical models such as large amount of calculation, high cost, inability to predict quickly, and reliance on feature engineering, and achieves fast and accurate prediction at low cost.

Description of the drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. , for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without exerting creative labor.

Figure 1 shows the implementation steps of a method for predicting significant wave height under ocean waves provided by the present invention;

Figure 2 is a specific flow diagram of Figure 1;

Figure 3 is a schematic diagram of the structure of the LSTM-Attention model of the ocean wave significant wave height prediction model;

Figure 4 is the error analysis index of Embodiment 2;

Figure 5 is a comparison chart of the significant wave height prediction effect in Embodiment 2.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Example 1:

Referring to Figures 1 to 3, the present invention provides a method for predicting significant wave height under ocean waves, which includes the following steps:

Step 1: Compose a data set based on historical NOAA buoy data; preprocess the data set to filter out missing values and outlier data, and then divide it into a training set and a test set. The data set includes wind direction, mean wind speed, peak wind speed, dominant wave period, mean wave period, wave direction, sea level pressure, air temperature, sea surface temperature and significant wave height. Among them, the wind direction, average wind speed, peak wind speed, dominant wave period, average wave period, wave direction, sea level pressure, air temperature and sea surface temperature data in the training set and test set are used as the input data of the training set and the test set respectively. Input data. The effective wave heights in the training set and test set are used as output data respectively.

When preprocessing a data set to filter out missing values and outlier data, data annotation, data cleaning, normalization and feature processing are required. The data set is based on time series and is divided into training set and test set according to the ratio of 3:1 of year data. That is, the 2019-2021 data is used for training and the 2022 data is used for testing.

Among them, the normalization processing calculation formula is:

Among them, X ^* is the standardized data, X is the original data, is the mean of the original data, and δ is the standard deviation of the original data. The processed feature data conforms to the standard normal distribution with a mean of 0 and a standard deviation of 1.

Step 2: Build a basic forecast model. Using the Tensorflow architecture, the LSTM layer is called from the keras library to build the LSTM model and then the Attention layer is added to obtain a prediction model that combines the LSTM neural network and the Attention mechanism.

Step 3: Use the training set in Step 1 to iteratively train the forecast model built in Step 2, and input the test set in Step 1 into the trained forecast model for testing; and perform denormalization processing according to The accuracy of the forecast model is verified, and the forecast model that meets the preset forecast accuracy standard evaluation is used as the significant wave height forecast model.

(3.1) The specific training methods of the ocean wave effective wave height prediction model include:

Input the input data of the training set into the LSTM neural network, predict according to the data change pattern at the previous moment, and output the LSTM prediction information; during the prediction process of the LSTM neural network, the characteristics of the time series data information are fully extracted.

The mathematical model of LSTM neural network is:

I _t =σ(X _t W _xi +H _t-1 W _hi +b _i )

F _t =σ(X _t W _xf +H _t-1 W _hf +b _f );

O _t =σ(X _t W _xo +H _t-1 W _ho +b _o );

In the formula: I _t , F _t , O _t represent the input gate, forgetting gate and output gate respectively; X _t represents the input data, W _ij represents the weight matrix of the output gate, and H _t-1 represents the hidden state of the previous time step. , b _i represents the bias term of the output gate, σ represents the activation function, C _t represents candidate memory cells and memory cells respectively, where/> Represents the multiplication of matrix elements.

(3.2) Input the LSTM prediction information into the added Attention layer. The Attention layer assigns probability weights to the output information of the LSTM neural network and dynamically adjusts the learning rate to improve the generalization of the effective wave height prediction model. .

The mathematical model of the Attention mechanism is:

e _t =σ(W _e x _t + _be )

In the formula: e _t , We _e , b _e , σ respectively represent the weight coefficient combination, trainable weight matrix, bias vector and activation function corresponding to each input data at the current moment;

Each attention weight coefficient is normalized through the softmax function to obtain the attention weight, where α _m,t is the attention weight value of the m-th feature, expressed as:

Recompute the input feature vector x _t as weighted Vector, expressed as:

In step three, the methods to verify the accuracy of the forecast model include:

(1) Input the test set into the trained forecast model, and obtain the forecast effective wave height through denormalization processing.

(2) Compare the predicted wave height with the effective wave height in the test set, and calculate the root mean square error, mean absolute error, mean absolute percentage error and goodness of fit as evaluation indicators to verify the accuracy of the model; the calculation formula is:

in, represents the deep learning forecast result; y represents the significant wave height in the NOAA data set;/> Represents the average significant wave height in the NOAA data set; N represents the number of observations; RMSE represents the root mean square error; MAE represents the mean absolute error; MAPE represents the absolute percentage error; R ² represents the goodness of fit.

Embodiment 2

As shown in Figure 4 and Figure 5, this embodiment uses the prediction method of effective wave height under sea waves in Embodiment 1, and uses the buoy data provided by NOAA as the experimental data set (https://www.ndbc.noaa.gov). The data buoys are 44013 and 44014. The geographical locations of the two buoys are as shown in the following table:

serial number Site name longitude latitude 1 44013 70.651°W 42.346°N 2 44014 74.842°W 36.609°N

The spatial resolution is 0.5° × 0.5°, and the data from 2019-2022 of buoys 44013 and 44014 are used to conduct effective wave height prediction experiments. The data set in the experiment is divided as shown in the following table:

serial number Site name training set length Test set length 1 44013 January 1, 2019 to December 31, 2021 January 1 to December 31, 2022 2 44014 January 1, 2019 to December 31, 2021 January 1 to October 3, 2022

The training set of 44013 and 44014 sites is from 0:00 on January 1, 2019 to 23:00 on December 31, 2021, and the test set of 44013 site is from 0:00 on January 1, 2022 to 23:00 on December 31; because the 44014 site is in Basically all the data is missing after 16:00 on October 3, 2022, so the test set of the 44014 site spans from 0:00 on January 1, 2022 to 16:00 on October 3, 2022. Since the data of NOAA buoy sites are missing for a certain period of time, the data sets of the buoy sites are first cleaned, and then the model is trained and verified. After cleaning, the data are normalized and then input into the significant wave height prediction model.

In order to further demonstrate the prediction effect of the proposed significant wave height prediction model, the same data set was input into the existing LSTM model and the applied significant wave height prediction model to compare the effects of the two models. Record the ocean wave significant wave height prediction model under application as LSTM-Attention (due to the length problem in the chart, it is abbreviated as LSTM-ATT in the chart); verify the model accuracy of the results predicted by the two models, and calculate the root mean square error , mean absolute error, mean absolute percentage error and goodness of fit. The results are shown in Figure 4. The applied wave significant wave height prediction model, LSTM-Attention, is better than the existing LSTM model at both sites 44013 and 44014. Better, the root mean square error can be reduced by 10.72%, the absolute percentage error can be reduced by 7.58%, the average absolute error can be reduced by 8.05%, and the goodness of fit can be improved by 5.91%. In order to intuitively see the difference between the two models, data visualization drawing is performed, as shown in Figure 4. In order to better compare the prediction effects of the two models and select a certain period of time to zoom in, the first row is a comparison chart of the prediction effects of the two models at the two sites. The second row is an enlarged view of the selected period. Figure 5(a) shows the prediction effect of the 44013 site. Combined with the enlarged view of Figure 5(a), it can be seen that the LSTM-Attention fitting effect is better. The existing The LSTM model here shows that the predicted value is smaller than the actual value; Figure 5(b) shows the prediction effect of the 44014 site. Also combined with its enlarged picture, it can be seen that LSTM-Attention fits better. The existing LSTM The model shows here that the predicted value is larger than the actual value. In summary, it can be seen that the applied significant wave height prediction model performs better than the existing LSTM model.

Those skilled in the art will understand that embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a use A device for realizing the functions specified in one process or multiple processes of the flowchart and/or one block or multiple blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction means, the instructions The device implements the functions specified in a process or processes of the flowchart and/or a block or blocks of the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operating steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby executing on the computer or other programmable device. Instructions provide steps for implementing the functions specified in a process or processes of a flowchart diagram and/or a block or blocks of a block diagram.

The above are only preferred embodiments of the present invention. It should be noted that those of ordinary skill in the art can also make several improvements and modifications without departing from the technical principles of the present invention. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (9)

1. A method for predicting effective wave height under ocean waves, including: Step 1: Compose a data set based on historical NOAA buoy data; preprocess the data set to filter out missing values and outlier data, and divide it into a training set and a test set; Step 2: Use the Tensorflow architecture, call the LSTM layer from the keras library to build the LSTM model and then add the Attention layer to obtain a forecast model that combines the LSTM neural network and the Attention mechanism; Step 3: Use the training set to iteratively train the forecast model, and input the test set into the trained forecast model for testing; and perform the accuracy of the forecast model according to the denormalization process. Verify that the forecast model that meets the preset forecast accuracy standard evaluation will be used as the effective wave height forecast model of sea waves; Step 4: After inputting the buoy data of the wave point to be measured into the output of the wave significant wave height prediction model, determine the significant wave height of the wave point to be measured. 2. The method for predicting effective wave height under ocean waves according to claim 1, characterized in that the data set includes wind direction, average wind speed, peak wind speed, dominant wave period, average wave period, wave direction, sea level pressure, Air temperature, sea surface temperature and significant wave height; wherein, the input data of the training set and test set respectively include wind direction, average wind speed, peak wind speed, dominant wave period, average wave period, wave direction, sea level pressure, air temperature and Sea surface temperature; the output data of the training set and the test set respectively include significant wave heights. 3. The method for predicting significant wave height under ocean waves according to claim 1, wherein the preprocessing of the data set in step one specifically includes: data annotation, data cleaning, normalization and feature processing. 4. The method for predicting significant wave height under ocean waves according to claim 1, characterized in that the data set is based on time series, and is divided into the training set and the test set according to the year data at a ratio of 3:1. 5. The prediction method of significant wave height under sea waves according to claim 2, characterized in that the specific training method of the sea wave significant wave height prediction model includes: Input the input data of the training set into the LSTM neural network, predict according to the data change pattern at the previous moment, and output the LSTM prediction information; The LSTM prediction information is input to the added Attention layer. The Attention layer assigns probability weights to the output information of the LSTM neural network and dynamically adjusts the learning rate to improve the generalization of the effective wave height prediction model. 6. The method for predicting significant wave height under ocean waves according to claim 2, characterized in that the mathematical model of the LSTM neural network is: I _t =σ(X _t W _xi +H _t-1 W _hi +b _i ) F _t =σ(X _t W _xf +H _t-1 W _hf +b _f ); O _t =σ(X _t W _xo +H _t-1 W _ho +b _o ); In the formula: I _t , F _t , O _t represent the input gate, forgetting gate and output gate respectively; X _t represents the input data, W _ij represents the weight matrix of the output gate, and H _t-1 represents the hidden state of the previous time step. , b _i represents the bias term of the output gate, σ represents the activation function, C _t represents candidate memory cells and memory cells respectively, where/> Represents the multiplication of matrix elements. 7. The method for predicting effective wave height under ocean waves according to claim 2, characterized in that the mathematical model of the Attention mechanism is: e _t =σ(W _e x _t + _be ) In the formula: e _t , We _e , b _e , σ respectively represent the weight coefficient combination, trainable weight matrix, bias vector and activation function corresponding to each input data at the current moment; Each attention weight coefficient is normalized through the softmax function to obtain the attention weight, where α _m,t is the attention weight value of the m-th feature, expressed as: Recompute the input feature vector x _t as weighted Vector, expressed as: 8. The method for predicting significant wave height under ocean waves according to claim 3, characterized in that the normalization processing calculation formula is: Among them, X ^* is the standardized data, X is the original data, is the mean of the original data, and δ is the standard deviation of the original data. The processed feature data conforms to the standard normal distribution with a mean of 0 and a standard deviation of 1. 9. The method for predicting significant wave height under ocean waves according to claim 1, characterized in that the method for verifying the accuracy of the forecast model in step three specifically includes: Input the test set into the trained forecast model, and obtain the predicted effective wave height through denormalization processing; Compare the predicted wave height with the effective wave height in the test set, and calculate the root mean square error, mean absolute error, mean absolute percentage error and goodness of fit as evaluation indicators to verify the accuracy of the model; The calculation formula is: in, represents the deep learning forecast result; y represents the significant wave height in the NOAA data set;/> Represents the average significant wave height in the NOAA data set; N represents the number of observations; RMSE represents the root mean square error; MAE represents the mean absolute error; MAPE represents the absolute percentage error; R ² represents the goodness of fit.