Storm surge forecasting method in combination with empirical orthogonal decomposition and deep learning

Abstract

The invention discloses a storm surge forecasting method combining empirical orthogonal decomposition and deep learning, and belongs to the field of ocean storm surge forecasting. The method comprises the following steps: a, determining a wind field area according to the position of a forecast ocean station and the like; b, collecting and arranging a 10-meter wind distance flat data matrix from the reanalysis data set, and collecting observed storm surge data from the ocean station; c, performing empirical orthogonal decomposition on the data matrix to obtain a main modal matrix and a main component matrix; d, taking a principal component in the principal component matrix as a forecast factor, taking the storm water increase data as a forecast quantity, and constructing a sample library; carrying out model training to obtain a long-short term memory neural network forecasting model; and e, collecting a 10-meter wind distance flat data matrix for future forecasting, calculating a principal component value, and substituting the principal component value into the forecasting model to obtain future storm surge data. According to the method, the precision of storm water increase forecasting can be improved, and compared with the actual observation value of storm water increase, the calculated root mean square error is low.

Description

Technical Field

[0001] The present invention relates to the field of ocean storm surge forecasting, and in particular to a storm surge forecasting method combining empirical orthogonal decomposition and deep learning. Background Art

[0002] Storm surge disaster refers to the abnormal rise of sea level caused by strong atmospheric disturbances, such as tropical cyclones and extratropical cyclones, which greatly exceeds the normal tide level in the affected sea areas. The severity of storm surge disasters is closely related to meteorological conditions. At present, there are three types of storm surge forecasts: the first type is the empirical formula method. This method uses experience to quantify different prediction effects into one or more parameters based on statistical analysis of observation data. This method can combine the maximum storm surge water level and typhoon characteristics such as air pressure and wind speed through empirical formulas. The second type is the fluid dynamics model method, which uses fluid dynamics and atmospheric driving to establish direct or coupled numerical models to predict storm surges. The fluid dynamics model method is the most common method at present. The disadvantage of this method is that it requires precise and detailed fluid dynamics equations, terrain data, typhoon models, boundary conditions and a large amount of calculations. The third type is the artificial neural network method. The artificial neural network method does not require the establishment of a complex data model, but only requires input data to adjust the internal connections of the network. Relying on its powerful self-learning, self-organization and adaptability, it can automatically extract the mapping rules of the combination through the learning of the system input and output sample pairs, and automatically adjust the network structure parameters to adapt to environmental changes. It has strong approximation and fitting capabilities. Based on this, artificial neural networks can effectively simulate the nonlinear relationship between storm surges and meteorological conditions and are widely used.

[0003] However, the existing storm surge neural network forecasting models have a relatively simple introduction of meteorological data. One method is to use single-point meteorological data observed near the forecast site, and the other is to use key parameters of typhoon observations (such as the typhoon's real-time path, maximum wind speed, central air pressure, and the radius of a seven-level wind circle). This method can only predict simple typhoon storm surges.

[0004] The meteorological conditions faced in actual storm surge forecasting are often more complex and are affected by a variety of weather systems, such as storm surge disasters caused by the combination of cold air and extratropical cyclones, and storm surge disasters caused by the combination of northward typhoons and ground high pressure. For storm surge disasters forced by complex weather systems, if only simple meteorological data are still introduced, the impact of meteorological conditions cannot be fully considered, and the accuracy of neural network simulation forecasts cannot be guaranteed. Summary of the invention

[0005] To solve the above technical problems, the present invention proposes a storm surge forecasting method combining empirical orthogonal decomposition and deep learning. This method takes the forecasting station as the center and selects a large range of meteorological field data covering the influencing weather systems to improve the model's forecasting ability. Moreover, the empirical orthogonal decomposition method is used to process the meteorological field data, which not only retains most of the meteorological field data information but also avoids introducing too many forecasting variables and causing overfitting of the neural network model.

[0006] The technical solution adopted by the present invention is as follows: A storm surge forecasting method combining empirical orthogonal decomposition and deep learning, comprising the following steps: a. Determine the wind field area according to the location of the forecasting ocean station and the characteristics of historical storm surge processes; b. Collect and collate the 10-meter wind anomaly data matrix from the reanalysis dataset according to the wind field area, and collect the observed storm surge data from the ocean station; c. Perform empirical orthogonal decomposition on the 10-meter wind anomaly data matrix to obtain the principal mode matrix and the principal component matrix; d. Use the principal components in the principal component matrix as forecasting factors and the storm surge data as forecasting variables to construct a sample library; use a long short-term memory neural network for model training to obtain a long short-term memory neural network forecasting model; e. Collect the future forecast 10-meter wind anomaly data matrix; f. Calculate the principal component values of the future forecast 10-meter wind anomaly data matrix; g. Substitute the calculated principal component values of the future forecast 10-meter wind anomaly data matrix into the long short-term memory neural network forecasting model obtained in step d to obtain the forecast future storm surge data.

[0007] The beneficial technical effects of the present invention are as follows: Storm surge forecasting not only depends on local meteorological field data but also requires a large range of data covering the entire weather system. This method takes the forecasting station as the center and selects meteorological field data covering the influencing weather systems, so that meteorological forcing signals can be more comprehensively introduced into the neural network model, which helps to improve the model's forecasting ability.

[0008] If a large number of meteorological forcing signals are directly introduced, it will lead to too many prediction factors in the model and too many neurons in the neural network model. This will not only increase the training time of the long short-term memory neural network model and the difficulty of convergence, but also easily cause overfitting of the factor model, resulting in good performance of the model in the training stage, while the performance drops sharply in the testing stage. This method uses empirical orthogonal decomposition to process meteorological data and extract the main modes of the meteorological field. The main modes can explain most of the variance of the meteorological field with only a small number of variables. Replacing the original meteorological field data with the main modes and introducing them into the neural network model not only reduces the training difficulty of the model, but also avoids the overfitting problem of the model. When making model predictions, the future wind field can be directly projected onto the main modes in the model training stage and then substituted into the neural network model, which can avoid performing empirical orthogonal decomposition operations again, that is, shortening the calculation time and having good prediction accuracy.

[0009] In summary, the present invention can significantly improve the accuracy of storm surge prediction. Compared with the actual observed values of storm surge, the calculated root mean square error is relatively low. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a schematic flow chart of the storm surge prediction method combining empirical orthogonal decomposition and deep learning of the present invention; Figure 2 is a schematic diagram of the spatial range of the wind field and the location of the Weifang Port Marine Station in a specific application example of the present invention; Figure 3 is a spatial distribution diagram of the 1st to 10th main modes in a specific application example of the present invention; Figure 4 is a spatial distribution diagram of the 11th to 20th main modes in a specific application example of the present invention; Figure 5 is a bar chart of the explained variance of the first 20 eigenvalues in a specific application example of the present invention; Figure 6 is a result diagram of the root mean square error of model training in a specific application example of the present invention; Figure 7 is a result diagram of the loss function of model training in a specific application example of the present invention; Figure 8 shows the storm surge curves obtained by the prediction method of the present invention, the existing numerical model prediction, and the observation. DETAILED DESCRIPTION OF THE INVENTION

[0011] When conducting the prediction of the storm surge neural network model, meteorological grid data centered on the forecast station and with a large enough scope should be introduced so as to take into account the forcing effects of different weather systems on the storm surge disaster. However, a new problem arises at this time. The large-scale meteorological grid data brings a large number of data variables. If directly introduced into the neural network model, on the one hand, it increases the training difficulty of the model, and on the other hand, it is likely to cause overfitting of the model. Based on this, this method proposes to use empirical orthogonal decomposition to process the meteorological data, extract the main modes of the meteorological grid data, and use a small number of main modes to replace the full-field meteorological data, thus avoiding the problem of neural network overfitting.

[0012] As Figure 1 shown, a storm surge forecast method combining empirical orthogonal decomposition and deep learning, which mainly includes two stages: the model training stage and the model forecasting stage.

[0013] In the model training stage, it includes the following steps: a. According to the location of the forecast ocean station and the characteristics of the historical storm surge process, determine the wind field area, and the area range should be large enough to cover the main weather systems affecting the storm surge of the forecast ocean station.

[0014] b. According to the wind field area, collect and collate the 10-meter wind anomaly data matrix from the reanalysis dataset, and collect the observed storm surge data . Among them: is the grid number of the wind field data; is the time series length in the model training stage; , represents the wind speed value of the th grid point at the th moment; , represents the storm surge value at the th moment.

[0015] c. Perform empirical orthogonal decomposition on to obtain the main mode matrix and the principal component matrix . Among them, , represents the value of the th parameter in the th mode;

[0016] Step c specifically includes: c1. Calculate the covariance matrix of : . represents the transpose of .

[0017] c2. Perform eigen-decomposition on the covariance matrix to obtain the eigenvalue matrix and the corresponding eigenvector matrix , satisfying . Among them, is a diagonal matrix, , which are the eigenvalues of the covariance matrix , arranged from largest to smallest. The magnitude of each eigenvalue reflects the contribution rate corresponding to that mode, i.e., the explained variance; Each column in is the eigenvector corresponding to

[0018] c3. The eigenvector matrix is the main mode matrix, i.e., .

[0019] c4. The principal component matrix is the projection of in the main mode matrix, i.e., denotes the transpose of

[0020] d. Take the first 20 principal components in the principal component matrix as the predictors, and take as the predictand to construct the sample library . Use the long short-term memory neural network method for model training to obtain the neural network prediction model , is the storm surge calculated by the neural network model, is the long short-term memory neural network model.

[0021] Subsequent model prediction is carried out. In the model prediction stage, it includes the following steps: e. Collect the 10-meter wind anomaly data matrix for future prediction, is the time series length in the model prediction stage.

[0022] f. Calculate the principal components of the future predicted wind anomaly data.

[0023] Specifically, it includes: f1. Select the first 20 modes in the main mode matrix .

[0024] f2. Calculate the projection value of in the first 20 main modes, which is the principal component value of the future predicted wind anomaly data.

[0025] g. Substitute into the long short-term memory neural network model trained in step d , that is , to obtain the future storm surge predicted by the neural network model .

[0026] h. Compare the observed future storm surge and the future storm surge predicted by the neural network model , and calculate the root mean square error. It can also be further compared with the root mean square error calculated by other prediction products to evaluate the prediction effect of the neural network model of the present invention.

[0027] The present invention will be further described below in conjunction with specific application examples.

[0028] Taking the establishment of a storm surge prediction model for the Weifang Port Marine Station as an example.

[0029] Model training stage: a. According to the location of the Weifang Port Marine Station, select the spatial range of the wind field as 116°~126°E, 33°~42°N ( Figure 2 ).

[0030] b. Download the hourly ERA5 reanalysis data from 2018 to 2022. There are 1149 groups of data in total. Select the first 60% of the data (851 groups) for model training, and select the last 40% of the data (568 groups) for model prediction. There are 1517 grid points within 116°~126°E, 33°~42°N. Considering that the wind speed is a vector and includes two variables, the number of grids is 3034. Select the 10-meter wind field data and perform anomaly processing on the data to obtain the matrix , where , .

[0031] Collect the hourly observed tide level data of the Weifang Port tide gauge station from 2018 to 2022, and perform harmonic analysis on the tide level data to obtain the storm surge data .

[0032] c. Perform empirical orthogonal decomposition on the first 60% of the wind field anomaly data to obtain the eigenvalues , the main mode matrix and the principal component matrix . Draw the spatial distribution map ( Figure 3 , Figure 4 ) and the bar chart ( Figure 5 ) of the first 20 eigenvalues, and calculate that the cumulative explained variance is 90.3%.

[0033] d. Select the first 20 principal components in the principal component matrix as predictors , select the first 60% of the storm surge data of Weifang Port To forecast the quantity, a sample library is constructed. The model is trained using the long short-term memory neural network method. The model training results are as follows: Figure 6 and Figure 7 As shown, the trained neural network and parameters are stored.

[0034] Model prediction stage: e. Select 40% of the wind field anomaly data matrix after sorting (in, ).

[0035] f. The matrix formed by the first 20 main modes , do matrix multiplication , and obtain the projection of 40% wind field anomaly data in the first 20 modes .

[0036] g. Substitute into the neural network forecast model ,Right now , and get the storm surge forecast for Weifang Port .

[0037] h. Comparison of storm surge observed in Weifang Port Storm surge predicted by neural network model The root mean square error of the calculation is 13.2cm. The root mean square error of the storm surge numerical model forecasting Weifang Port during the same period is 16.9cm, which is 21.9% lower than that of the neural network model established by this method. Figure 8 This is a storm surge curve for an example in the forecast stage.

[0038] The parts not mentioned in the above methods can be realized by adopting or drawing on existing technologies.

[0039] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning, characterized in that It includes the following steps: a. Determine the wind field area according to the location of the forecast ocean station and the characteristics of historical storm surge processes; b. Collect and collate the 10-meter wind anomaly data matrix from the reanalysis dataset according to the wind field area, and collect the observed storm surge data from the ocean station; c. Perform empirical orthogonal decomposition on the 10-meter wind anomaly data matrix to obtain the main mode matrix and the principal component matrix; d. Use the principal components in the principal component matrix as forecast factors and the storm surge data as the forecast quantity to construct a sample library; use a long short-term memory neural network for model training to obtain a long short-term memory neural network forecast model; e. Collect the future forecast 10-meter wind anomaly data matrix; f. Calculate the principal component values of the future forecast 10-meter wind anomaly data matrix; g. Substitute the calculated principal component values of the future forecast 10-meter wind anomaly data matrix into the long short-term memory neural network forecast model obtained in step d to obtain the forecast future storm surge data.

2. A storm surge prediction method combining empirical orthogonal decomposition and deep learning according to claim 1, characterized in that In step a: The range of the determined wind field area should be able to cover the main weather systems affecting the storm surge at the forecast ocean station.

3. A storm surge prediction method combining empirical orthogonal decomposition and deep learning according to claim 1, characterized in that, In step b: 10-meter wind anomaly data matrix , where is the number of grids of wind field data, is the time series length in the model training stage; , represents the wind speed value of the th grid point at the th moment; Collect the observed storm surge data from ocean stations , , indicating the storm surge value at the th moment.

4. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 3, characterized in that In step c: For the matrix perform empirical orthogonal decomposition to obtain the main modal matrix and the principal component matrix ; where , represents the value of the i-th parameter in the j-th mode; , represents the value of the j-th mode at the t-th moment.

5. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 4, characterized in that, Step c specifically includes: c1. Calculate the covariance matrix of the 10-meter wind anomaly data matrix : ; denotes transpose; c2. Perform eigen-decomposition on the covariance matrix to obtain the eigenvalue matrix and the corresponding eigenvector matrix , satisfying ; where is a diagonal matrix, , and is the eigenvalue of the covariance matrix , arranged in descending order. The magnitude of each eigenvalue reflects the contribution rate corresponding to that mode; Each column in is the eigenvector corresponding to The eigenvector matrix is the main modal matrix, that is ; c4, principal component matrix is the projection in the main modal matrix, i.e., ; denotes the transpose of.

6. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 5, characterized in that In step d: Take the first 20 principal components in the principal component matrix as the predictors, and take the storm surge data as the predicted variable to construct a sample library ; ; The obtained long short-term memory neural network prediction model , is the storm surge data calculated by the long short-term memory neural network prediction model, which is the long short-term memory neural network prediction model.

7. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 6, characterized in that, In step e: The 10-meter wind anomaly data matrix of the future forecast is , is the length of the time series in the model forecast stage.

8. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 7, characterized in that, Step f specifically includes: f1. Select the first 20 modes from the main modal matrix ; ; f2, Calculation matrix Projection values in the first 20 main modes , which are the principal component values of the 10-meter wind anomaly data matrix for future forecasts.

9. A storm surge forecasting method combining empirical orthogonal decomposition and deep learning according to claim 8, characterized in that, In step g: Substitute the principal component values of the 10-meter wind anomaly data matrix for future forecasts into the long short-term memory neural network prediction model , that is , and obtain the future storm surge data predicted by the long short-term memory neural network prediction model .

Images