A wave forecasting method based on an optimal combination model

Abstract

The present invention provides a wave forecasting method based on an optimal combination model, which relates to the technical field of wave forecasting and includes the following steps: S1: Establish a wave database for the target area; S2: Form a one-dimensional array with the wave height values of each sample point in the wave database at different times; S3: Optimize the one-dimensional array through the whale optimization algorithm to obtain the parameters of the optimal combination model; S4: Obtain the optimal decomposed subsequence array; S5: Obtain the optimal combination model; S6: Use the gradient descent algorithm to optimize the parameters in the fully connected layer and convolutional kernel of the optimal combination model to obtain a prediction model; S7: Substitute the current wave height value of the target wave into the prediction model and output the predicted wave value of the target wave. Through the combination of variational mode decomposition, convolutional kernel, and fully connected neural network, the present invention achieves the effect of predicting the height of strongly nonlinear sea waves with fewer input conditions in a local area.

Description

Technical Field

[0001] The present invention relates to the technical field of wave forecasting, and in particular, to a wave forecasting method based on an optimal combination model. Background Art

[0002] The mainstream method of wave forecasting is the third-generation ocean wave model (WAN) method obtained by integrating and calculating based on the energy balance equation. In international professional wave forecasting tasks, derivative algorithms based on this method dominate. This method is based on the energy balance equation, so it has certain limitations.

[0003] In a local area, applying the method based on the energy balance equation to predict the significant wave height of ocean waves requires a large number of input boundary conditions. Generally speaking, the boundary conditions that need to be determined for wave prediction at least include the wind speed, temperature, air pressure in this area, and other ocean currents, tides, etc. caused by the earth's movement. In a local area, determining so many boundary conditions requires more manpower, material resources, and financial resources. Even when the boundary conditions are known, it is difficult to accurately predict the significant wave height of ocean waves at each point in this area with high precision. Under strong non-linear and non-stationary conditions, the decrease in accuracy brought by the method based on the energy balance equation is very obvious. This is because under strong non-linear conditions, the measurement of boundary conditions is more difficult and the changes are very drastic, making it difficult to achieve stationarity. Therefore, when the boundary conditions are unstable or even uncertain, the wave height predicted by the energy balance equation also has great uncertainty. Reflected in practice, it means that the wave prediction accuracy of this method is very low under strong typhoon and high sea state conditions. If the accuracy is improved, the effective prediction time will be sharply shortened.

[0004] In addition, the measured ocean waves are always contaminated by noise. The sources of noise pollution are complex and it is difficult to analyze the noise components. The existence of noise will greatly reduce the accuracy of the ocean wave database, making the input boundary conditions of the prediction model deviate. The resulting decrease in prediction accuracy cannot be achieved by replacing the prediction model. To solve such a problem, only a denoising algorithm can be added to the forecasting algorithm, or an algorithm with denoising ability can be used to eliminate the noise.

[0005] In summary, it is necessary to invent a method that can denoise in a local area and predict the height of strong non-linear ocean waves with fewer input conditions. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to propose a wave forecasting method based on an optimal combination model to solve the technical problem that the wave forecasting results are inaccurate due to the large number of input boundary conditions and the difficulty in accurately predicting the boundary conditions in the existing wave forecasting methods.

[0007] The technical means adopted by the present invention are as follows:

[0008] A wave forecasting method based on an optimal combination model, comprising the following steps:

[0009] S1: Establish a wave database for the target area, where the wave database includes the sea wave height values of each sample point in the target area within a certain time range;

[0010] S2: Combine the sea wave height values of each sample point in the wave database at different times into a one-dimensional array;

[0011] S3: Optimize the one-dimensional array through the whale optimization algorithm to obtain the parameters of the optimal combination model;

[0012] S4: Use the variational mode decomposition method to decompose the one-dimensional array and the parameters of the optimal combination model to obtain the optimal decomposed subsequence array;

[0013] S5: Perform convolution calculation on the optimal decomposed subsequence array, and perform pooling on the calculated values and then input them into the fully connected layer to obtain the optimal combination model;

[0014] S6: Use the gradient descent algorithm to optimize the parameters in the fully connected layer and the convolution kernel in the optimal combination model to obtain the prediction model;

[0015] S7: Substitute the current wave height value of the target wave into the prediction model and output the predicted wave value of the target wave.

[0016] Further, the selection rules for the sample points include: in the target area, buoys are used to represent the sample points, and the interval between the sample points is greater than or equal to the minimum activity range of the buoys; the sample points cover the entire target area; when measuring the significant wave height of the sample points, a complete strong non-linear and non-stationary period is included; the significant wave height of the sample points is measured at equal time intervals; the equal time interval is greater than or equal to 15 minutes.

[0017] Further, S4 includes the following steps:

[0018] Obtain the initialization matrix; the initialization matrix includes the decomposed subsequence matrix Center frequency matrix Lagrange multiplier matrix

[0019]

[0020]

[0021]

[0022] Where: n is the number of iterations, k is the number of modal decompositions, the value range of i is from 1 to k, and the value range of j is from 1 to n. is a one-dimensional array;

[0023] Substitute the one-dimensional array into the initialization matrix for iteration. When the iteration stop condition is met, export the subsequence matrix; the iteration stop condition is:

[0024]

[0025] where ε is the restricted bandwidth. When the stop condition is met, export the matrix as the subsequence matrix;

[0026] Divide the subsequence matrix into k subsequence arrays according to different sequences.

[0027] Furthermore, S5 includes:

[0028] For any subsequence array In the l-th convolutional layer, perform the following convolutional calculation:

[0029]

[0030] Where: is the output of this convolutional layer; is the bias of the m-th neuron in the l-th layer, is the kernel function from the p-th neuron in the l-th layer to the m-th neuron in the l-th layer. conv1D(.,.) is used to perform one-dimensional convolution without zero padding;

[0031]

[0032] where f(.) represents the activation function, and the activation function adopts one of the sigmoid and tanh machine learning activation functions, represents the predicted wave height value of this subsequence.

[0033] Furthermore, S6 includes:

[0034] Calculate the loss function E , ,

[0039] , q ,

[0038] ,

[0036] , , , , ,

[0035] ,

[0037] , , p , :

[0035]

[0036] Where: t q is the true measured wave height;

[0037] Solve the gradients of each parameter:

[0038]

[0039]

[0040] Update the parameters using the gradient descent formula:

[0041]

[0042]

[0043] where represents the parameter value at the t-th iteration;

[0044] When the parameter optimization is completed, the gradient becomes zero, and the parameter value will not change with the increase of the number of iterations, and the optimization stops immediately;

[0045] After completing the parameter optimization, the establishment of the prediction model is randomly completed.

[0046] The present invention also provides a storage medium, which includes a stored program. When the program runs, it executes the wave prediction method based on the optimal combination model described in any one of the above.

[0047] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor runs through the computer program to execute the wave prediction method based on the optimal combination model described in any one of the above.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] 1. The present invention uses the whale optimization algorithm to find the optimal parameter combination of variational mode decomposition. So that the decomposed subsequences do not have mode aliasing, and the frequency domain characteristics of each subsequence are obvious. Making the calculation efficiency of the overall model higher.

[0050] 2. The present invention uses the variational mode decomposition method for decomposition, so that the subsequences after decomposing the wave sequence all have different dominant frequencies, filtering out the noise of the contaminated data.

[0051] 3. The present invention optimizes the parameter combination of variational mode decomposition through the whale optimization algorithm, making the calculation efficiency of the overall model higher.

[0052] 4. The present invention combines variational mode decomposition with convolutional kernels and fully connected neural networks to achieve the effect of predicting the height of strongly nonlinear ocean waves with fewer input conditions in a local area. Description of the Drawings

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0054] Figure 1 This is the flowchart of the method of the present invention.

[0055] Figure 2 This is an example diagram of the buoy layout in the target sea area of the present invention.

[0056] Figure 3 This is the flowchart of the variational mode decomposition of the present invention.

[0057] Figure 4 This is the flowchart of the optimized mode decomposition of the present invention.

[0058] Figure 5 This is the structural topology diagram of the optimal combination model of the present invention.

[0059] Figure 6 This is an example diagram of the data table header of the present invention. Detailed implementation manners

[0060] In order to enable those skilled in the art of the present technology to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0062] As Figures 1 - 5 shown, the present invention provides a wave forecasting method based on an optimal combination model, including the following steps:

[0063] Build a wave database for the target area. Specifically, the database contains the wave height values at various positions in the target area within a certain time range. The selection of points is as shown in the figure: Based on covering the entire target area and arranged in a certain rule, as many points as possible should be selected, but the interval between points cannot be less than the minimum interval of the buoy. When measuring the significant wave height of these points, it is best to include a complete strong non-linear and non-stationary period. Measurements should be taken at equal time intervals, and the interval time should be as short as possible. However, it cannot be less than 15 minutes. The database is presented in tabular form, as Figure 6 shown.

[0064] Form a one-dimensional array from the measured values of each point in the database at different times

[0065] Optimize the parameters of VMD decomposition to obtain the optimal decomposition result, as Figure 4 shown.

[0066] The whale optimization algorithm is a meta-heuristic optimization algorithm based on the iterative evolution search of the humpback whale population. By simulating the unique bubble-net attack hunting behavior of humpback whales, it realizes the optimization of complex optimization problems. The specific optimization mechanism includes two parts: the local exploitation stage and the global exploration stage:

[0067] In the local exploitation stage, the whale assumes that the current prey position is the target position and continuously approaches the prey with a 50% probability between the shrinking encirclement mechanism and the spiral update mechanism to obtain the optimal solution. Its mathematical model is as follows:

[0068]

[0069] Where: X * is the current target position; X is the whale position; A is a coefficient whose absolute value is less than or equal to 1; D' = X * (t) - X(t) represents the distance from the whale to the prey; b is a constant defined as the shape of the logarithmic spiral; l is a random number in (-1, 1) to make the spiral shape irregular for better searching for the optimal solution.

[0070] In the global exploration stage, the whale uses the positions of other whales as references and continuously updates its own position for random search. Its mathematical model can be expressed as:

[0071] X(t + 1) = X rand (t) - A·D

[0072] Where the absolute value of A is greater than 1; X rand is the randomly selected whale position.

[0073] The objective function is selected as permutation entropy V PEN, and its derivation process is as follows:

[0074] The phase space of the component sequence {X(i), i = 1, 2, …, N} is reconstructed into matrix Y:

[0075]

[0076] where K, τ, and d represent the embedding dimension, time delay, and number of components, respectively.

[0077] In matrix Y, each reconstructed component is rearranged in ascending order, and the column indices of the positions of the elements are: j1, j2, …, j K .

[0078] Define the symbol sequence: For any matrix obtained by reconstructing a time series, a set of symbol sequences can be obtained for each row in Y:

[0079] S = {j1, j2, …, j K}

[0080] Calculate the probabilities {P1, P2, …, P K} of the occurrence of each symbol sequence. The entropy of the time series X(i) is:

[0081]

[0082] Use variational mode decomposition to decompose and optimize the first 70% of the data of the one-dimensional array. As Figure 3 shown, define three matrices, which represent the subsequence matrix, center frequency matrix, and Lagrange multiplier matrix after decomposition, respectively. Define two positive integers n and k, which represent the number of iterations and the number of mode decompositions, respectively. The value range of i is 1 - k; the value range of j is 1 - n.

[0083] Let λ be randomly initialized, that is, fill the entire matrix with random numbers.

[0084]

[0085]

[0086]

[0087] Among them, represents the original sequence, that is, the one-dimensional array obtained in the previous step.

[0088] The iteration stop condition is:

[0089]

[0090] Among them, ε is the restricted bandwidth. When the stop condition is satisfied, the matrix is derived, which is the matrix composed of the decomposed subsequences. The matrix composed of the subsequences is divided into u1, u2, u3... u k according to different sequences, a total of k subsequence arrays.

[0091] Perform convolution calculations on each of the decomposed arrays and build a fully connected layer. As Figure 5 shown.

[0092] For any subsequence array in the l-th convolutional layer, perform the following convolution calculation:

[0093]

[0094] Where: is the output of this convolutional layer; the bias of the m-th neuron in the l-th layer, is the kernel function from the p-th neuron in the l-th layer to the m-th neuron in the l-th layer. conv1D(.,.) is used to perform one-dimensional convolution without zero padding.

[0095]

[0096] Among them, f(.) represents the activation function. The activation function can be selected from classical machine learning activation functions such as sigmoid and tanh. represents the predicted wave height value of this subsequence.

[0097] Set the gradient descent algorithm to optimize the parameters in the fully connected layer and the convolutional kernel. After training, a prediction model is obtained.

[0098] Calculate the loss function E p :

[0099]

[0100] Where: t q is the true measured wave height.

[0101] Solve the gradient of each parameter:

[0102]

[0103]

[0104] Update the parameters using the gradient descent formula:

[0105]

[0106]

[0107] wherein represents the parameter value at the t-th iteration step.

[0108] Set the number of convolutional layers L to 3; set the number of iteration steps as much as possible, such as 500; set the learning rate α to 0.001;

[0109] When the parameter optimization is completed, the gradient will automatically become zero, and the parameter value will not change with the increase of the number of iterations, and the optimization will stop immediately.

[0110] After completing the parameter optimization, the establishment of the prediction model is randomly completed. All unknowns in the above formula are clear, and the output wave value can be predicted by substituting the input wave value.

[0111] Use the last 30% of the one-dimensional array for comparison and verification to test the reliability of the model.

[0112] After the model is verified to be reliable, the same operation can be performed on other points in the target area using the same method, and a model library for this area can be obtained. The significant wave height values of each point in this area can be predicted using the model library.

[0113] The present invention also provides a storage medium, the storage medium includes a stored program, wherein, when the program runs, it executes a wave prediction method based on an optimal combination model.

[0114] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes a wave prediction method based on an optimal combination model through the running of the computer program.

[0115] Variational mode decomposition (VMD) is a method for analyzing non-stationary data. For a non-stationary one-dimensional signal, through VMD decomposition, the data can be decomposed into a series of sub-signals and a trend signal. Different sub-signals have different dominant frequencies, and the trend signal can represent the trend of the non-stationary signal. Compared with other signal decomposition algorithms, VMD can not only decompose the non-stationary characteristics of the signal, but also has advantages such as avoiding mode mixing and good noise robustness. Convolutional neural network is a common method for extracting spatial features. Through the calculation of "corresponding multiplication and summation" by several convolutional kernels at a certain step in the data, the features of the data are matched with the features of the convolutional kernels. By continuously changing the data of the convolutional kernels and using the gradient descent algorithm to optimize the data of the convolution sum, feature extraction is finally realized. 1D-CNN also has advantages such as low algorithm complexity and fast calculation speed.

[0116] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0117] In the above embodiments of the present invention, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0118] In the several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0119] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0120] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0121] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the respective embodiments of the present invention. And the aforementioned storage medium includes: USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical disks and other various media that can store program codes.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wave forecasting method based on an optimal combination model, characterized in that, It includes the following steps: S1: Establish a wave database for the target area, where the wave database includes the wave height values of each sample point in the target area within a certain time range; S2: Combine the wave height values of each sample point in the wave database at different times into a one-dimensional array; S3: Optimize the one-dimensional array through the whale optimization algorithm to obtain the parameters of the optimal combination model; S4: Use the variational mode decomposition method to decompose the one-dimensional array and the parameters of the optimal combination model to obtain the optimal decomposed subsequence array; S5: Perform convolution calculation on the optimal decomposed subsequence array, and perform pooling on the calculated values and then input them into the fully connected layer to obtain the optimal combination model; S6: Use the gradient descent algorithm to optimize the parameters in the fully connected layer and the convolution kernel in the optimal combination model to obtain the prediction model; S7: Substitute the current wave height value of the target wave into the prediction model and output the predicted wave value of the target wave.

2. The wave forecasting method based on the optimal combination model according to claim 1, wherein, The selection rules for the sample points include: in the target area, buoys are used to represent the sample points, and the interval between the sample points is greater than or equal to the minimum activity range of the buoy; the sample points cover the entire target area; when measuring the significant wave height of the sample points, it includes a complete strong nonlinear and non-stationary period; the significant wave height of the sample points is measured at equal time intervals; the equal time interval is greater than or equal to 15 minutes.

3. The wave forecasting method based on the optimal combination model according to claim 1, wherein S4 includes the following steps: Obtain an initialization matrix; the initialization matrix includes a decomposed subsequence matrix Center frequency matrix Lagrange multiplier matrix Where: n is the number of iterations, k is the number of modal decompositions, the value range of i is from 1 to k, and the value range of j is from 1 to n. is a one-dimensional array; Substitute the one-dimensional array into the initialization matrix for iteration. When the iteration stop condition is met, export the subsequence matrix; the iteration stop condition is: where ε is the restricted bandwidth, and when the bandwidth stop condition is satisfied, the matrix is exported as the subsequence matrix; Divide the subsequence matrix into k subsequence arrays according to different sequences.

4. The wave prediction method based on the optimal combination model according to claim 1, characterized in that S5 It includes: For any subsequence array In the l-th convolutional layer, the following convolution calculation is performed: Wherein: is the output of the convolutional layer; is the bias of the m-th neuron in the l-th layer, is the kernel function from the p-th neuron in the l-th layer to the m-th neuron in the l-th layer, and conv1D(.,.) is used to perform one-dimensional convolution without zero-padding; Among them, f(.) represents the activation function, and the activation function adopts one of the sigmoid and tanh machine learning activation functions. represents the predicted wave height value of this subsequence.

5. The wave forecasting method based on the optimal combination model according to claim 1, characterized in that S6 It includes: Calculate the loss function E p : where: t q is the true measured wave height; Solve the gradients of each parameter: Update the parameters using the gradient descent formula: where represents the parameter value at the t-th iteration; When the parameter optimization is completed, the gradient is set to zero, and the parameter values will not change with the increase of the iteration times, and the optimization stops immediately; After the parameter optimization is completed, the establishment of the prediction model is randomly completed.

6. A storage medium, characterized in that, The storage medium includes a stored program, wherein when the program runs, it executes the wave prediction method based on the optimal combination model described in any one of claims 1 to 5.

7. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor runs through the computer program to execute the wave prediction method based on the optimal combination model described in any one of claims 1 to 5.

Images