A wave forecast model generation method, device, medium and equipment

Abstract

The present invention relates to the field of wave forecasting technology, and discloses a wave forecasting model generation method, device, medium and equipment, which can convert target wave surface rise time series data to obtain target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point in the target test period during the wave tank test in the physical wave tank; the target wave surface envelope is input as a training data into a pre-trained wave forecasting model for training to obtain a trained wave forecasting model. The present invention can obtain a target wave surface envelope based on a wave tank test, and use the target wave surface envelope to train the pre-trained wave forecasting model to obtain a trained wave forecasting model, and the trained wave forecasting model can be used for intelligent wave forecasting to realize intelligent wave forecasting.

Description

Technical Field

[0001] The present invention relates to the technical field of wave forecasting, and in particular to a wave forecasting model generating method, device, medium and equipment. Background Art

[0002] Extreme sea conditions in deep sea environments have challenged the safe operation of various deep sea equipment. Forecasting sea waves, that is, forecasting the wave conditions in a certain area of ​​the sea in the future, is of great significance to the selection of various operation windows for my country's marine engineering, as well as the performance improvement and safety assurance of marine equipment.

[0003] Currently, the relevant technology can be used by technicians to perform wave forecasting based on experience. However, the relevant technology cannot effectively realize intelligent wave forecasting. Summary of the invention

[0004] The present invention provides a wave forecast model generation method, device, medium and equipment, which are used to solve the defect that intelligent wave forecast cannot be effectively realized in the related technology, and realize intelligent wave forecast.

[0005] In a first aspect, the present invention provides a wave forecast model generation method, the method comprising:

[0006] The target wave surface rise time series data is converted to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank;

[0007] The target wave envelope is input as a training data into a pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period;

[0008] A loss function value is determined according to the predicted wave envelope and the corresponding true wave envelope, and the parameters in the pre-trained wave forecast model are updated according to the loss function value to obtain a trained wave forecast model.

[0009] Optionally, the pre-trained wave forecast model includes a nonlinear Schrödinger equation, and the nonlinear Schrödinger equation includes an envelope prediction function to be solved;

[0010] The pre-trained wave forecast model is used to input the training data into the nonlinear Schrödinger equation to solve the envelope prediction function to be solved, and obtain the solved envelope prediction function as the envelope prediction sub-model.

[0011] Optionally, the nonlinear Schrödinger equation is:

[0012] ;

[0013] in, is an imaginary unit; x is the spatial coordinate, used to identify the location of the water tank; t It is a time for testing; is the envelope prediction function to be solved, and The corresponding envelope; for x The marked water tank position point at the time of the test t The wave surface rises; for The phase of is the carrier wave number, is the carrier frequency.

[0014] Optionally, determining the loss function value according to the predicted envelope envelope and the corresponding true envelope envelope includes:

[0015] Determine a first loss function value according to the predicted wave envelope and the true wave envelope, and determine a second loss function value according to the predicted wave envelope; wherein the first loss function value is used to characterize the deviation between the predicted wave envelope and the true wave envelope, and the second loss function value is used to characterize the deviation between the predicted wave envelope and the wave evolution data described by the envelope forecast sub-model;

[0016] The loss function value is determined based on the first loss function value and the second loss function value.

[0017] Optionally, determining a first loss function value according to the predicted envelope envelope and the true envelope envelope includes:

[0018] Converting the wave surface rise time series data of the downstream position in the second test period to obtain the real wave surface envelope;

[0019] The true envelope envelope and the predicted envelope envelope are input into a data sample loss function to perform loss function calculation to obtain the first loss function value.

[0020] Optionally, determining a second loss function value according to the predicted envelope envelopment includes:

[0021] Performing partial derivative calculation on the predicted envelope to obtain corresponding partial derivative calculation results;

[0022] The partial derivative calculation result is input into the control equation solving loss function to perform loss function calculation to obtain the second loss function value.

[0023] Optionally, determining the loss function value based on the first loss function value and the second loss function value includes:

[0024] Calculating a first product of the first loss function value and a first weight coefficient, and calculating a second product of the second loss function value and a second weight coefficient;

[0025] A sum of the first product and the second product is calculated, and the sum is determined as the loss function value.

[0026] In a second aspect, the present invention provides a wave forecast model generating device, comprising:

[0027] A conversion unit, used to convert the target wave surface rise time series data to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank;

[0028] An input unit, inputting the target wave envelope as a training data into a pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period;

[0029] A determination unit, which determines a loss function value according to the predicted wave envelope and the corresponding true wave envelope;

[0030] An updating unit updates the parameters in the pre-trained wave forecast model according to the loss function value to obtain a trained wave forecast model.

[0031] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the wave forecast model generation method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the wave forecast model generation method of the first aspect or any corresponding embodiment thereof.

[0033] The wave forecast model generation method, device, medium and equipment provided by the present invention can convert the target wave surface rise time series data to obtain the target wave envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test time period during the wave tank test in the physical wave tank; the target wave envelope is input as a training data into the pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test time period, and according to the downstream position point to be predicted and the second test time period; the loss function value is determined according to the predicted wave envelope and the corresponding true wave envelope, and the parameters in the pre-trained wave forecast model are updated according to the loss function value to obtain a trained wave forecast model. The present invention can use the target wave envelope obtained based on the wave tank test to train the pre-trained wave forecast model to obtain a trained wave forecast model, and the trained wave forecast model can be used for intelligent wave forecasting to realize intelligent wave forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flow chart of a wave forecast model generation method provided by an embodiment of the present invention;

[0036] Figure 2 A flow chart of another wave forecast model generation method provided by an embodiment of the present invention;

[0037] Figure 3 A schematic structural diagram of a wave forecast model generating device provided by an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In the related technologies, the linear prediction model has high computational efficiency but does not take into account the strong nonlinearity of waves and thus has low accuracy; the high-order spectral method has greatly improved the accuracy of wave forecasting by taking into account high-order nonlinearities, but its low computational efficiency cannot meet the requirements of rapid forecasting; the prediction method that solves the nonlinear Schrödinger equation has the problem of poor applicability; data-driven methods such as neural networks have poor interpretability and unsatisfactory generalization performance because their models do not contain physical information.

[0041] This embodiment can be committed to developing a method for improving the accuracy of wave forecasting by using physical information neural networks based on full consideration of the physical mechanism and characteristics of wave evolution, and realizing accurate intelligent forecasting of waves by using a method for solving the nonlinear Schrödinger equation.

[0042] Combine the following Figure 1-Figure 2 The wave forecast model generation method of the present invention is described.

[0043] like Figure 1 As shown, this embodiment proposes a first wave forecast model generation method, which may include the following steps:

[0044] S101. Convert the target wave surface rise time series data to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank.

[0045] The target test period is a certain period during the wave tank test in the physical wave tank.

[0046] The water tank position point may be a spatial position point in a physical wave tank where the wave height changes.

[0047] The target wave surface rise time series data is the wave surface rise time series of the corresponding tank position point during the target test period in the wave tank test. It should be noted that wave surface rise refers to the change in the height of the wave crest relative to the mean sea level during the wave propagation process.

[0048] Specifically, this embodiment can carry out a wave tank test in a physical wave tank, and longitudinally set a plurality of groups of resistive wave height meters at different tank positions along the propagation evolution direction of the wave. Through wave tank tests with different test parameters, the wave surface rise time series at different tank positions are obtained, providing basic data and multiple training data for training the wave prediction model. It can be understood that the wave surface rise time series obtained at each tank position can be used as a target wave surface rise time series data.

[0049] Specifically, this embodiment can transform the target wave surface rise time series data into the corresponding wave envelope, i.e., the target wave envelope, through Hilbert transform, so as to effectively extract the wave envelope information for subsequent processing and analysis.

[0050] Optionally, the wave rise at a certain point in the wave tank test at a certain time can be expressed as To express. With Envelope The relationship between To express it as:

[0051] ;

[0052] ;

[0053] Further, Representing the wave surface The phase of the carrier wave number and frequency , and the location of the sink and time It can be expressed as:

[0054] .

[0055] S102. Input the target wave envelope as a training data into the pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period.

[0056] The pre-trained wave forecast model may have a certain wave forecast capability. In this embodiment, the pre-trained wave forecast model may be trained to improve its wave forecast capability until the wave forecast performance of the pre-trained wave forecast model meets the requirements. It should be noted that this embodiment does not limit the specific type of the pre-trained wave forecast model. For example, the pre-trained wave forecast model may be a pre-trained neural network model with a certain wave forecast capability.

[0057] The upstream position point corresponds to the downstream position point. The upstream position point may be a spatial position point in the physical wave tank that is located upstream of the tank, and the downstream position point may be a spatial position point in the physical wave tank that is located downstream of the tank.

[0058] Specifically, the first test period and the second test period may be a period during the wave tank test in the physical wave tank. The start time of the second test period may be after the start time of the first test period.

[0059] Specifically, the envelope prediction sub-model has the ability to predict wave envelope. Specifically, it can predict and generate the wave envelope corresponding to the wave surface rise time series data of the input upstream position point in the second test time period, as well as the input downstream position point to be predicted and the second test time period, that is, predict the wave envelope.

[0060] Optionally, the pre-trained wave prediction model includes a nonlinear Schrödinger equation, and the nonlinear Schrödinger equation includes an envelope prediction function to be solved;

[0061] The pre-trained wave forecast model is used to input the training data into the nonlinear Schrödinger equation to solve the envelope prediction function to be solved, and the solved envelope prediction function is obtained as the envelope prediction sub-model.

[0062] Alternatively, the nonlinear Schrödinger equation is:

[0063] ;

[0064] in, is an imaginary unit; x is the spatial coordinate, used to identify the location of the water tank; t It is a time for testing; is the envelope prediction function to be solved, and The corresponding envelope; for x The marked water tank position point at the time of the test t The wave surface rises; for The phase of is the carrier wave number, is the carrier frequency.

[0065] Specifically, the pre-trained wave forecast model can be a neural network model with embedded physical information constructed in this embodiment, which can perform calculations and solutions and wave forecasts based on the nonlinear Schrödinger equation. The neural network model can cover the physical characteristics of waves and the wave evolution mechanism described by the nonlinear Schrödinger equation. This embodiment can improve the interpretability and generalization performance of the wave forecast model by introducing physical information into the neural network model and training the neural network model to generate a trained wave forecast model.

[0066] It should be noted that the target wave envelope can be used as training data to train the pre-trained wave forecast model, update the parameters in the pre-trained wave forecast model, and improve the wave forecast performance of the pre-trained wave forecast model until a wave forecast model with wave forecast performance that meets the requirements is obtained, that is, a trained wave forecast model.

[0067] S103, determining a loss function value according to the predicted envelope and the corresponding true envelope.

[0068] Specifically, this embodiment can calculate the loss function value according to the predicted wavelet envelope, the corresponding true wavelet envelope and the loss function.

[0069] Optionally, in other wave forecast model generation methods proposed in this embodiment, step S103 may include:

[0070] Determine a first loss function value according to the predicted wave envelope and the true wave envelope, and determine a second loss function value according to the predicted wave envelope; wherein the first loss function value is used to characterize the deviation between the predicted wave envelope and the true wave envelope, and the second loss function value is used to characterize the deviation between the predicted wave envelope and the wave evolution data described by the envelope forecast sub-model;

[0071] A loss function value is determined based on the first loss function value and the second loss function value.

[0072] It should be noted that, in this embodiment, a loss function can be defined based on the above-mentioned neural network, which can specifically include a data sample loss function and a control equation solution loss function. The data sample loss function can be used to measure the deviation between the neural network prediction value and the wave tank test data. The control equation solution loss function can be used to measure the deviation between the neural network prediction value and the wave evolution described by the envelope prediction sub-model. The combination of these two parts of the loss function can ensure the prediction accuracy of the wave prediction model, and ensure that the above-mentioned neural network prediction process of the wave complies with the physical nonlinear evolution mechanism and control equation.

[0073] Optionally, the above-mentioned determining the first loss function value according to the predicted wave envelope and the true wave envelope includes:

[0074] Transforming the wave surface rise time series data at the downstream position during the second test period to obtain the real wave envelope;

[0075] The true wave envelope and the predicted wave envelope are input into the data sample loss function to calculate the loss function and obtain the first loss function value.

[0076] Specifically, in the wave flume test, the present embodiment can record the wave surface rise time series data at the downstream position in the second test period, convert the wave surface rise time series data into a corresponding wave envelope, and use it as the real wave envelope. Afterwards, the present embodiment can input the real wave envelope and the predicted wave envelope into the data sample loss function to calculate the first loss function value, and measure the deviation between the predicted wave envelope and the real wave envelope.

[0077] Among them, the data sample loss function can be:

[0078] .

[0079] It represents the total number of sample points, specifically the total number of wave surface rise data in the wave surface rise time series data at the downstream position during the second test period. Ldata express represents the index of the sample point, represents the predicted wave envelope, represents the true wave envelope.

[0080] Optionally, the determining of the second loss function value according to the predicted envelope envelope includes:

[0081] Perform partial derivative calculation on the predicted wave envelope to obtain the corresponding partial derivative calculation results;

[0082] The partial derivative calculation result is input into the control equation to solve the loss function to calculate the loss function and obtain the second loss function value.

[0083] Among them, the loss function for solving the control equation can be:

[0084] .

[0085] in, N represents the total number of sample points, n represents the index of the sample point, represents the predicted wave envelope, represents the imaginary unit, Representing the wave surface The phase of represents the carrier wave number, Indicates the carrier frequency. Represents a function pair The partial derivative of .

[0086] Specifically, in this embodiment, the sum of the first loss function value and the second loss function value can be directly determined as the loss function value.

[0087] Optionally, this embodiment may also weight the first loss function value and the second loss function value according to actual conditions, and then perform weighted calculation to obtain the loss function value.

[0088] Optionally, the determining of the loss function value based on the first loss function value and the second loss function value includes:

[0089] Calculating a first product of a first loss function value and a first weight coefficient, and calculating a second product of a second loss function value and a second weight coefficient;

[0090] A sum of the first product and the second product is calculated, and the sum is determined as a loss function value.

[0091] The first product is the product of the first loss function value and the first weight coefficient, and the second product is the product of the second loss function value and the second weight coefficient.

[0092] Specifically, the specific sizes of the first weight coefficient and the second weight coefficient can be set by technical personnel according to actual conditions, and this embodiment does not limit this.

[0093] S104. Update the parameters in the pre-trained wave forecast model according to the loss function value to obtain a trained wave forecast model.

[0094] Specifically, this embodiment can perform supervised learning based on the loss function value and the optimization algorithm, and update the parameters in the pre-trained wave forecast model until the loss function value meets certain requirements (such as being less than a set loss function limit) or the number of training iterations is greater than a set value.

[0095] Specifically, the present embodiment can compare the loss function value with the set loss function limit value, and when the loss function value is greater than the loss function limit value, the parameters in the pre-trained wave forecast model can be updated. Afterwards, the present embodiment can convert other target wave surface rise time series data into corresponding target wave surface envelopes, and input them as new training data into the current pre-trained wave forecast model for training to obtain a new loss function value. When the new loss function value is greater than the loss function limit value, the parameters in the current pre-trained wave forecast model are updated again until the latest loss function value meets certain requirements or the number of training iterations is greater than the set value. At this time, the pre-trained wave forecast model can meet the wave forecast performance requirements and can be used as a trained wave forecast model.

[0096] In this embodiment, during the model training process, the model parameters can be continuously updated to find the hyperparameters that minimize the loss function value.

[0097] Specifically, the present embodiment can use a trained wave forecast model to forecast sea waves. The present embodiment can obtain the time series data of the wave surface rise at the upstream position within a certain period of time, and convert it into a corresponding wave envelope, input the wave envelope into the trained wave forecast model, and determine the downstream position and time to be predicted. The trained wave forecast model can forecast waves based on the wave envelope corresponding to the downstream position at the time according to the input wave envelope, and obtain the predicted wave envelope generated and output by the trained wave forecast model, and then perform inverse transformation on the predicted wave envelope, convert it into the corresponding wave surface rise time series data and determine it as the wave surface rise time series data of the downstream position at the time, so as to achieve accurate forecasting of waves.

[0098] Optionally, in the process of training the pre-trained wave forecast model, this embodiment can select the wave forecast model composed of the neural network parameters with the highest test accuracy as the trained wave forecast model according to the performance of the validation set and the test set.

[0099] It should be noted that this embodiment uses the test data obtained by measuring in the wave test pool, and establishes an embedded physical information neural network for solving the nonlinear Schrödinger equation based on the wave prediction model of the nonlinear Schrödinger equation, so as to predict the nonlinear evolution process of waves with time and space, which can improve the accuracy and efficiency of wave forecasting.

[0100] The method of using a physical information neural network to solve the nonlinear Schrödinger equation to achieve wave forecasting does not involve the adjustment of the theory of the nonlinear Schrödinger equation, but is aimed at improving the problem of unsatisfactory wave forecasting accuracy and efficiency caused by the limitations of the theoretical assumptions and the low efficiency of solving the equation. By combining the physical information neural network, the present embodiment can effectively improve the accuracy of wave forecasting by the nonlinear Schrödinger equation, and improve the forecasting efficiency, thereby realizing intelligent wave forecasting.

[0101] In order to solve the above-mentioned nonlinear Schrödinger equation, which has problems of insufficient applicability and decreased solution accuracy due to high-order nonlinearity and wave bandwidth limitation, this embodiment can also adopt the equation parameter identification method of physical information neural network, correct the nonlinear parameters of the original equation on the basis of considering the physical empirical laws and the characteristics of waves with different bandwidths, and establish a more general and more explanatory deep learning model.

[0102] This embodiment can also be applied to consider the inverse problem of parameter identification of the nonlinear Schrödinger equation and construct a deep learning neural network. Its basic design principle is to approximate the solution of the partial differential equation by training a fully connected neural network, and use the automatic differentiation technology of the neural network to obtain the partial derivatives of the solution of the equation of each order, so as to add the partial differential equation and the initial and boundary value residuals as regularization terms into the loss function, and train the neural network by designing the architecture of the loss function.

[0103] It should also be noted that, after obtaining the trained wave forecast model, the present embodiment can predict the wave envelope of the downstream position point in a certain period or time according to the wave rise time series data of the upstream position point in a certain period. Specifically, the present embodiment can first obtain the wave rise time series data of the upstream position in a certain period, convert the wave rise time series data into the corresponding wave envelope, input the wave envelope, the position information of the downstream position point and the period or time to be predicted into the trained wave forecast model to generate the predicted wave envelope corresponding to the downstream position point, and convert the predicted wave envelope into the corresponding wave rise time series data or wave rise, that is, the wave rise time series data of the downstream position point in the period to be predicted or the wave rise at the time to be predicted.

[0104] The wave prediction model generation method proposed in this embodiment can convert the target wave surface rise time series data to obtain the target wave envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point in the target test period during the wave tank test in the physical wave tank; the target wave envelope envelope is input as a training data into the pre-trained wave prediction model, so that the pre-trained wave prediction model generates an envelope prediction sub-model according to the training data; the envelope prediction sub-model is used to generate the corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be predicted and the second test period; the loss function value is determined according to the predicted wave envelope and the corresponding true wave envelope, and the parameters in the pre-trained wave prediction model are updated according to the loss function value to obtain a trained wave prediction model. In this embodiment, the target wave envelope envelope obtained based on the wave tank test can be used to train the pre-trained wave prediction model to obtain a trained wave prediction model, and the trained wave prediction model can be used for intelligent wave forecasting to realize intelligent wave forecasting.

[0105] like Figure 2 As shown, this embodiment proposes another wave forecast model generation method. In this method, this embodiment can carry out a wave tank test 1 based on physical wave tanks, wave machines, wave breakers and other equipment under different wave parameters, and longitudinally set a number of groups of resistive wave height meters at different spatial positions along the propagation evolution direction of the wave to obtain the change of wave surface rise over time at different spatial positions, that is, to obtain multiple target wave surface rise time series data.

[0106] A neural network model is established as a pre-trained wave forecast model, which consists of an input layer 2, a hidden layer 3 and an output layer 4. x , t and B is a model-related parameter. In this embodiment, the time series data of the wavefront rise required can be passed to the input layer 2, and then transmitted through the hidden layer 3, and finally mapped to the output layer 4 to generate an envelope prediction sub-model. The wave envelope calculated using the envelope prediction sub-model is the predicted wave envelope. According to the wave envelope calculated by the neural network, based on the mechanism of automatic calculation of gradients of the neural network, the partial derivative calculation 5 of the predicted wave envelope is performed.

[0107] Among them, the partial derivative calculation 5 can be obtained including , and The partial derivative calculation results included.

[0108] in, Figure 2 The physical information output by the physical wave tank test 1 can be the wave envelope corresponding to the control result, that is, the real wave envelope.

[0109] Afterwards, this embodiment can wrap the real wave envelope and Figure 2 middle I The predicted envelope represented by is input into the data sample loss function 6 to calculate the corresponding first loss function value, which is used to characterize the deviation between the neural network calculation and the actual test data. The partial derivative calculation result is input into the control equation solution loss function 9, and based on the mathematical form of the control equation solution loss function 9, that is, the above-mentioned nonlinear Schrödinger equation, the control equation solution loss function 7 is calculated to add the control equation as a regularization term into the loss function. The total loss function value 8 during the neural network training process is equal to the sum of the first loss function value calculated by the data sample loss function 6 and the second loss function value calculated by the control equation solution loss function 7.

[0110] This embodiment can be based on the calculated loss function value and use the optimization algorithm to perform supervised learning, and update the parameters of the neural network model 10 until the loss function value meets certain requirements or the number of training iterations is greater than the set value. This embodiment can determine whether the number of iterations is greater than the set value. If the judgment result is Yes That is, when the iteration number is greater than the set value, the PINN training is determined to be completed. No That is, when the iteration number is not greater than the set value, the parameters of the neural network model continue to be updated.

[0111] In this embodiment, different spatial coordinates and times to be solved can be input into the input layer 2 of the network according to the trained neural network model, and after calculation by the neural network, a rapid intelligent prediction result of wave surface rise within the entire calculation time and space range can be finally achieved.

[0112] The wave forecast model generation method proposed in this embodiment can use the test data obtained by measuring in the wave test pool to establish an embedded physical information neural network for solving the nonlinear Schrödinger equation, train a wave forecast model based on the nonlinear Schrödinger equation, and intelligently forecast the nonlinear evolution process of waves with time and space. While realizing intelligent forecasting, it can effectively ensure the accuracy and efficiency of wave forecasting.

[0113] and Figure 1 The method shown corresponds to Figure 3 As shown, this embodiment provides a wave forecast model generating device, comprising:

[0114] The conversion unit 101 is used to convert the target wave surface rise time series data to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank;

[0115] The input unit 102 inputs the target wave envelope as a training data into the pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period;

[0116] A determination unit 103 determines a loss function value according to a predicted envelope and a corresponding true envelope;

[0117] The updating unit 104 updates the parameters in the pre-trained wave forecast model according to the loss function value to obtain a trained wave forecast model.

[0118] It should be noted that the processing of the conversion unit 101, the input unit 102, the determination unit 103 and the updating unit 104 and the beneficial effects thereof can be respectively referred to in Figure 1 The relevant descriptions of steps S101 to S104 are not repeated here.

[0119] Optionally, the pre-trained wave prediction model includes a nonlinear Schrödinger equation, and the nonlinear Schrödinger equation includes an envelope prediction function to be solved;

[0120] The pre-trained wave forecast model is used to input the training data into the nonlinear Schrödinger equation to solve the envelope prediction function to be solved, and the solved envelope prediction function is obtained as the envelope prediction sub-model.

[0121] Alternatively, the nonlinear Schrödinger equation is:

[0122] ;

[0123] in, is an imaginary unit; x is the spatial coordinate, used to identify the location of the water tank; t It is a time for testing; is the envelope prediction function to be solved, and The corresponding envelope; for x The marked water tank position point at the time of the test t The wave surface rises; for The phase of is the carrier wave number, is the carrier frequency.

[0124] Optionally, the determining unit 103 is further configured to:

[0125] Determine a first loss function value according to the predicted wave envelope and the true wave envelope, and determine a second loss function value according to the predicted wave envelope; wherein the first loss function value is used to characterize the deviation between the predicted wave envelope and the true wave envelope, and the second loss function value is used to characterize the deviation between the predicted wave envelope and the wave evolution data described by the envelope forecast sub-model;

[0126] A loss function value is determined based on the first loss function value and the second loss function value.

[0127] Optionally, the determining unit 103 is further configured to:

[0128] Transforming the wave surface rise time series data at the downstream position during the second test period to obtain the real wave envelope;

[0129] The true wave envelope and the predicted wave envelope are input into the data sample loss function to calculate the loss function and obtain the first loss function value.

[0130] Optionally, the determining unit 103 is further configured to:

[0131] Perform partial derivative calculation on the predicted wave envelope to obtain the corresponding partial derivative calculation results;

[0132] The partial derivative calculation result is input into the control equation to solve the loss function to calculate the loss function and obtain the second loss function value.

[0133] Optionally, the determining unit 103 is further configured to:

[0134] Calculating a first product of a first loss function value and a first weight coefficient, and calculating a second product of a second loss function value and a second weight coefficient;

[0135] A sum of the first product and the second product is calculated, and the sum is determined as a loss function value.

[0136] The wave prediction model generation device proposed in this embodiment can convert the target wave surface rise time series data to obtain the target wave envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point in the target test period during the wave tank test in the physical wave tank; the target wave envelope envelope is input as a training data into the pre-trained wave prediction model, so that the pre-trained wave prediction model generates an envelope prediction sub-model according to the training data; the envelope prediction sub-model is used to generate the corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be predicted and the second test period; the loss function value is determined according to the predicted wave envelope and the corresponding true wave envelope, and the parameters in the pre-trained wave prediction model are updated according to the loss function value to obtain the trained wave prediction model. In this embodiment, the target wave envelope obtained based on the wave tank test can be used to train the pre-trained wave prediction model to obtain the trained wave prediction model, and the trained wave prediction model can be used for intelligent wave forecasting to realize intelligent wave forecasting. This embodiment can obtain a wave envelope used as training data based on a wave tank test, input the wave envelope into a pre-trained model for wave forecasting, and train the pre-trained model to obtain a trained wave forecast model. The trained wave forecast model can be used for wave forecasting to achieve intelligent wave forecasting.

[0137] The wave forecast model generating device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0138] The embodiment of the present invention also provides a computer device having the above Figure 3 The wave forecast model generating device shown.

[0139] See also Figure 4, a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses for communication, and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 4 A processor 10 is taken as an example.

[0140] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0141] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0142] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function. The data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0143] The memory 20 may include a volatile memory, such as a random access memory. The memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive. The memory 20 may also include a combination of the above-mentioned types of memory.

[0144] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0145] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0146] 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wave forecast model generation method, characterized in that: include: The target wave surface rise time series data is converted to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank; The target wave envelope is input as a training data into a pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period; Determining a loss function value according to the predicted wave envelope and the corresponding true wave envelope, and updating parameters in the pre-trained wave forecast model according to the loss function value to obtain a trained wave forecast model; Wherein, the pre-trained wave prediction model includes a nonlinear Schrödinger equation, and the nonlinear Schrödinger equation includes an envelope prediction function to be solved; The pre-trained wave forecast model is used to input the training data into the nonlinear Schrödinger equation to solve the envelope prediction function to be solved, and obtain the solved envelope prediction function as the envelope prediction sub-model.

2. The method according to claim 1, characterized in that The nonlinear Schrödinger equation is: ； in, is an imaginary unit; x is the spatial coordinate, used to identify the location of the water tank; t It is a time for testing; is the envelope prediction function to be solved, and The corresponding envelope; for x The marked water tank position point at the time of the test t The wave surface rises; for The phase of is the carrier wave number, is the carrier frequency.

3. The method according to any one of claims 1 to 2, characterized in that The determining of the loss function value according to the predicted envelope and the corresponding true envelope comprises: Determine a first loss function value according to the predicted wave envelope and the true wave envelope, and determine a second loss function value according to the predicted wave envelope; wherein the first loss function value is used to characterize the deviation between the predicted wave envelope and the true wave envelope, and the second loss function value is used to characterize the deviation between the predicted wave envelope and the wave evolution data described by the envelope forecast sub-model; The loss function value is determined based on the first loss function value and the second loss function value.

4. The method according to claim 3, characterized in that The determining of a first loss function value according to the predicted envelope envelope and the true envelope envelope comprises: Converting the wave surface rise time series data of the downstream position in the second test period to obtain the real wave surface envelope; The true envelope envelope and the predicted envelope envelope are input into a data sample loss function to perform loss function calculation to obtain the first loss function value.

5. The method according to claim 3, characterized in that: The determining of the second loss function value according to the predicted envelope envelope comprises: Performing partial derivative calculation on the predicted envelope to obtain corresponding partial derivative calculation results; The partial derivative calculation result is input into the control equation solving loss function to perform loss function calculation to obtain the second loss function value.

6. The method according to claim 3, characterized in that The determining the loss function value based on the first loss function value and the second loss function value includes: Calculating a first product of the first loss function value and a first weight coefficient, and calculating a second product of the second loss function value and a second weight coefficient; A sum of the first product and the second product is calculated, and the sum is determined as the loss function value.

7. A wave forecast model generating device, characterized in that: include: A conversion unit, used to convert the target wave surface rise time series data to obtain the target wave surface envelope; wherein the target wave surface rise time series data is the wave surface rise time series data of the tank position point within the target test period during the wave tank test in the physical wave tank; An input unit, inputting the target wave envelope as a training data into a pre-trained wave forecast model, so that the pre-trained wave forecast model generates an envelope forecast sub-model according to the training data; the envelope forecast sub-model is used to generate a corresponding predicted wave envelope according to the wave envelope corresponding to the wave surface rise time series data of the upstream position point in the first test period, and according to the downstream position point to be forecasted and the second test period; A determination unit, which determines a loss function value according to the predicted wave envelope and the corresponding true wave envelope; an updating unit, which updates the parameters in the pre-trained wave forecast model according to the loss function value to obtain a trained wave forecast model; Wherein, the pre-trained wave prediction model includes a nonlinear Schrödinger equation, and the nonlinear Schrödinger equation includes an envelope prediction function to be solved; The pre-trained wave forecast model is used to input the training data into the nonlinear Schrödinger equation to solve the envelope prediction function to be solved, and obtain the solved envelope prediction function as the envelope prediction sub-model.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the wave forecast model generation method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the wave forecast model generation method according to any one of claims 1 to 6.

Images