A wave directional spectrum recognition model generation method, device, equipment and medium

Abstract

The present invention relates to the field of wave directional spectrum recognition, and discloses a wave directional spectrum recognition model generation method, device, equipment and medium. The present invention can determine multiple wave surface rise time series data and spatial distribution matrix in the process of wave simulation test in a numerical wave pool, and the spatial distribution matrix is ​​used to record the spatial distribution of multiple measurement positions and non-measurement positions in the numerical wave pool; generate multiple signal spectra according to the multiple wave surface rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position; use the multiple signal spectra and the spatial distribution matrix as a whole as training data to train a pre-trained wave directional spectrum recognition model to obtain a trained wave directional spectrum recognition model. The wave directional spectrum recognition model trained by the present invention can be used for wave directional spectrum recognition, thereby effectively realizing intelligent recognition of wave directional spectrum recognition.

Description

Technical Field

[0001] The present invention relates to the field of wave directional spectrum recognition, and in particular to a wave directional spectrum recognition model generation method, device, equipment and medium. Background Art

[0002] In recent years, extreme sea conditions have frequently occurred in deep-sea environments, posing serious challenges to the safe operation of various deep-sea equipment.

[0003] Accurate forecasting of marine waves, especially the accurate and rapid identification and estimation of wave directional spectra, is of great significance for the selection of various operation windows for my country's marine engineering projects, as well as the environmental wave perception, performance improvement, and safety assurance of marine equipment.

[0004] However, the relevant technology cannot effectively realize the intelligent recognition of wave direction spectrum. Summary of the invention

[0005] The present invention provides a wave direction spectrum recognition model generation method, device, equipment and medium, which are used to effectively realize intelligent recognition of wave direction spectrum.

[0006] In a first aspect, the present invention provides a method for generating a wave direction spectrum recognition model, comprising:

[0007] Determine a plurality of wave surface rise time series data corresponding to a plurality of measurement positions in the numerical wave pool, and determine a spatial distribution matrix; wherein the wave surface rise time series data is obtained by measuring the wave surface rise at the corresponding measurement positions during a wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool;

[0008] Generate multiple signal spectra according to the multiple wavefront rise time series data; wherein each of the signal spectra corresponds to the measurement position, and each of the signal spectra is marked with a position identifier for identifying the corresponding measurement position;

[0009] The multiple signal spectra and the spatial distribution matrix are taken as a training data as a whole, and the pre-trained wave direction spectrum recognition model is trained using the training data to obtain a trained wave direction spectrum recognition model.

[0010] Optionally, determining the spatial distribution matrix includes:

[0011] Gridding the numerical wave pool to obtain a plurality of grids;

[0012] Generate a matrix to be assigned corresponding to the plurality of grids, wherein the matrix to be assigned includes elements to be assigned corresponding to the grids;

[0013] Determine the grid where each of the measurement positions is located and use them as a first grid respectively; and determine the grid where each of the non-measurement positions is located and use them as a second grid respectively;

[0014] In the matrix to be assigned a value, each element to be assigned a value corresponding to the first grid is assigned a value of 1, and each element to be assigned a value corresponding to the second grid is assigned a value of 0, so as to obtain the spatial distribution matrix.

[0015] Optionally, generating a plurality of signal spectra according to the plurality of wavefront rise time series data includes:

[0016] For any of the wavefront rise time series data, generate a corresponding frequency spectrum and a first position identifier according to the wavefront rise time series data, wherein the first position identifier is used to identify the measurement position corresponding to the wavefront rise time series data, and mark the first position identifier on the frequency spectrum;

[0017] For any two of the wavefront rise time series data, generate a corresponding cross spectrum and a second position identifier according to the two wavefront rise time series data, the second position identifier is used to identify the measurement position corresponding to the two wavefront rise time series data, and mark the second position identifier on the cross spectrum;

[0018] Normalizing all the frequency spectra and all the cross spectra to obtain a corresponding plurality of normalized spectra, each of the normalized spectra being marked with the first position marker or the second position marker;

[0019] Each of the normalized spectra is determined as one of the signal spectra.

[0020] Optionally, when the pre-trained model is a pre-trained generative adversarial network, the pre-trained model includes a frequency wave direction estimation network and a discriminant network;

[0021] The using the training data to train the pre-trained wave direction spectrum recognition model comprises:

[0022] Inputting the training data into the frequency wave direction estimation network to identify the wave direction spectrum and obtain an estimated wave direction spectrum;

[0023] Determining a real wave direction spectrum corresponding to the estimated wave direction spectrum according to the wave simulation test;

[0024] Inputting the estimated wave direction spectrum and the real wave direction spectrum into the discrimination network for calculation to obtain the credibility corresponding to the estimated wave direction spectrum;

[0025] A loss function value is determined according to the estimated wave direction spectrum, the true wave direction spectrum and the credibility, and parameters in the pre-trained wave direction spectrum recognition model are updated according to the loss function value.

[0026] Optionally, the frequency wave direction estimation network includes: a frequency domain identification module and a space identification module;

[0027] The step of inputting the training data into the frequency wave direction estimation network to identify the wave direction spectrum and obtain the estimated wave direction spectrum comprises:

[0028] Inputting the multiple signal spectra in the training data into the frequency domain identification module, so that the frequency domain identification module performs energy distribution identification according to the multiple signal spectra and the position identifier marked on each of the signal spectra to obtain a target feature vector; wherein the target feature vector includes distribution characteristics of wave energy at different frequencies;

[0029] Input the target feature vector and the spatial distribution matrix into the spatial identification module for directional distribution identification to obtain a target matrix; wherein the target matrix includes the distribution characteristics of wave energy in different directions within each frequency domain;

[0030] The target matrix is ​​determined as the estimated wave direction spectrum.

[0031] Optionally, determining the loss function value according to the estimated wave direction spectrum, the real wave direction spectrum and the credibility includes:

[0032] Inputting the estimated wave direction spectrum and the real wave direction spectrum into a first loss function for calculation, to obtain a first loss function value output by the first loss function;

[0033] Inputting the credibility into a second loss function for calculation to obtain a second loss function value output by the second loss function;

[0034] The first loss function value and the second loss function value are added to obtain the loss function value.

[0035] Optionally, the first loss function value and the second loss function value are respectively:

[0036]

[0037] Among them, L MSE is the first loss function value, s and represent the true wave direction spectrum and the estimated wave direction spectrum respectively, L d is the second loss function value, For the credibility.

[0038] In a second aspect, the present invention provides a wave direction spectrum recognition model generation device, comprising:

[0039] A determination unit, used to determine a plurality of wave surface rise time series data corresponding to a plurality of measurement positions in a numerical wave pool, and to determine a spatial distribution matrix; wherein the wave surface rise time series data are obtained by measuring the wave surface rise at the corresponding measurement positions during a wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool;

[0040] A generating unit, configured to generate a plurality of signal spectra according to the plurality of wavefront rise time series data; wherein each of the signal spectra corresponds to the measurement position, and each of the signal spectra is marked with a position identifier for identifying the corresponding measurement position;

[0041] As a unit, used to treat the plurality of signal spectra and the spatial distribution matrix as a training data as a whole;

[0042] A training unit is used to train the pre-trained wave direction spectrum recognition model using the training data to obtain a trained wave direction spectrum recognition model.

[0043] 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 direction spectrum recognition model generation method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0044] 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 direction spectrum recognition model generation method of the first aspect or any corresponding embodiment thereof.

[0045] The wave direction spectrum recognition model generation method, device, equipment and medium provided by the present invention can determine multiple wave surface rise time series data corresponding to multiple measurement positions in a numerical wave pool, and determine a spatial distribution matrix; wherein the wave surface rise time series data is obtained by measuring the wave surface rise at the corresponding measurement position during a wave simulation test in the numerical wave pool; the spatial distribution matrix is ​​used to record the spatial distribution of multiple measurement positions and non-measurement positions in the numerical wave pool; multiple signal spectra are generated according to the multiple wave surface rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position; the multiple signal spectra and the spatial distribution matrix are used as a training data as a whole, and the training data is used to train the pre-trained wave direction spectrum recognition model to obtain a trained wave direction spectrum recognition model. The trained wave direction spectrum recognition model of the present invention can be used for wave direction spectrum recognition, thereby effectively realizing intelligent recognition of wave direction spectrum recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] 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.

[0047] Figure 1 A flow chart of a method for generating a wave direction spectrum recognition model provided by an embodiment of the present invention;

[0048] Figure 2 A flow chart of another wave direction spectrum recognition model generation method provided by an embodiment of the present invention;

[0049] Figure 3 A schematic diagram of a wave direction spectrum in matrix form provided by an embodiment of the present invention;

[0050] Figure 4 A schematic diagram of a general form of wave direction spectrum provided by an embodiment of the present invention;

[0051] Figure 5 A schematic diagram of the structure of a wave direction spectrum recognition model generation device provided by an embodiment of the present invention;

[0052] Figure 6 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] 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.

[0054] Combine the following Figure 1-Figure 4 The wave direction spectrum recognition model generation method of the present invention is described.

[0055] like Figure 1 As shown, this embodiment proposes a first wave direction spectrum recognition model generation method, which may include the following steps:

[0056] S101, determining a plurality of wave surface rise time series data corresponding one to one to a plurality of measurement positions in a numerical wave pool, and determining a spatial distribution matrix; wherein the wave surface rise time series data are obtained by measuring the wave surface rise at corresponding measurement positions during a wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool.

[0057] It should be noted that, in this embodiment, wave surface rise sensors can be set at different positions in the numerical wave pool, and during the wave simulation test on the numerical wave pool, each wave surface rise sensor is instructed to measure the wave surface rise at its position at multiple identical times in the same period of time to obtain the wave surface rise time series data at different positions.

[0058] The measurement position may be a position in the numerical wave pool where a wave surface rise sensor is disposed.

[0059] The non-measurement position may be a position in the numerical wave pool where no wave surface rise sensor is provided.

[0060] like Figure 2 As shown, in this embodiment, five wave height meters 2, namely wave height meters #1 to #5, can be respectively arranged at five different positions of a numerical wave pool 1 including a wave-breaking beach. During a wave simulation test, namely a numerical pool test, in the numerical wave pool, each wave height meter can perform wave surface rise measurement of the same frequency in the same period of time, and obtain wave surface rise curves at times such as T1, T2, T3 and T4, that is, a plurality of wave surface rise time series data 3 can be obtained.

[0061] Optionally, the above-mentioned determination of the spatial distribution matrix may include:

[0062] The numerical wave pool is meshed to obtain multiple grids;

[0063] Generate a matrix to be assigned corresponding to the plurality of grids, wherein the matrix to be assigned includes elements to be assigned corresponding to the grids;

[0064] Determine a grid where each measurement position is located and use it as a first grid respectively; and determine a grid where each non-measurement position is located and use it as a second grid respectively;

[0065] In the matrix to be assigned, each element to be assigned corresponding to the first grid is assigned a value of 1, and each element to be assigned corresponding to the second grid is assigned a value of 0, so as to obtain a spatial distribution matrix.

[0066] like Figure 2 As shown, in this embodiment, grid division can be performed in the numerical wave pool to determine the grid where each measurement position, i.e., each wave height meter, is located, and the wave height meter spatial distribution coordinates 8 are obtained, and then the corresponding wave height meter spatial distribution matrix 9 is generated by assigning 0 or 1.

[0067] Specifically, this embodiment can carry out a three-dimensional short-peak wave simulation in a numerical wave pool based on a high-order nonlinear wave simulation method, and arrange no less than four groups of wave height meters at different measurement positions in the numerical wave pool to monitor the wave surface rise time series at the position.

[0068] Specifically, this embodiment can be based on the wave history at different measurement positions obtained in the numerical wave pool, and the distribution of different wave height meter spatial positions can be represented by a spatial distribution 0-1 matrix.

[0069] S102, generating a plurality of signal spectra according to a plurality of wavefront rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position.

[0070] Optionally, step S102 may include:

[0071] For any wavefront rise time series data, a corresponding frequency spectrum and a first position identifier are generated according to the wavefront rise time series data, the first position identifier is used to identify the measurement position corresponding to the wavefront rise time series data, and the first position identifier is marked on the frequency spectrum;

[0072] For any two wavefront rise time series data, a corresponding cross spectrum and a second position identifier are generated according to the two wavefront rise time series data, the second position identifier is used to identify the measurement position corresponding to the two wavefront rise time series data, and the second position identifier is marked on the cross spectrum;

[0073] Normalizing all the frequency spectra and all the cross spectra to obtain a corresponding plurality of normalized spectra, each normalized spectrum being marked with a first position marker or a second position marker;

[0074] Each normalized spectrum is determined as a signal spectrum.

[0075] Specifically, this embodiment can generate a corresponding frequency spectrum according to each wavefront rise time series data, and generate a corresponding cross spectrum according to every two wavefront rise time series data.

[0076] like Figure 2 As shown, this embodiment can generate wave spectrum / cross spectrum results, that is, multiple signal spectra, based on multiple wave surface rise time series data 3.

[0077] S103, taking the multiple signal spectra and spatial distribution matrices as a whole as a training data.

[0078] Specifically, in this embodiment, the generated multiple signal spectra and spatial distribution matrices can be used as training data to train the pre-trained wave direction spectrum recognition model.

[0079] It can be understood that this embodiment can consider different combinations of significant wave heights, spectral peak periods and directional distribution parameters during the numerical wave simulation, i.e., wave simulation test, to construct a short-peak wave field test sample including a large number of sea state parameters, and consider the distribution form combination of different wave height meter spatial positions to obtain multiple training data.

[0080] It should be noted that, in this embodiment, multiple training data can be determined by changing the test parameters of the wave simulation test, the measurement position, or performing the test multiple times.

[0081] S104: Use the training data to train the pre-trained wave direction spectrum recognition model to obtain a trained wave direction spectrum recognition model.

[0082] The pre-trained wave direction spectrum recognition model may be a model with certain wave direction spectrum recognition performance. This embodiment does not limit the model type.

[0083] Specifically, in this embodiment, the training data can be used to train the pre-trained wave direction spectrum recognition model to update its model parameters.

[0084] It should be noted that, in this embodiment, a plurality of training data obtained from wave simulation tests can be used to train the pre-trained wave direction spectrum recognition model, and optimize its wave direction spectrum recognition performance until its performance meets the requirements, so as to obtain a trained wave direction spectrum recognition model.

[0085] Specifically, this embodiment can iteratively train the pre-trained wave direction spectrum recognition model for multiple rounds to gradually optimize and generate various parameters of the pre-trained wave direction spectrum recognition model. During the training process, the model parameters are continuously updated to find various hyperparameters that minimize the loss function, thereby obtaining a trained wave direction spectrum recognition model.

[0086] The wave direction spectrum recognition model generation method proposed in this embodiment can determine multiple wave surface rise time series data corresponding to multiple measurement positions in the numerical wave pool, and determine the spatial distribution matrix; wherein the wave surface rise time series data is obtained by measuring the wave surface rise at the corresponding measurement position during the wave simulation test in the numerical wave pool; the spatial distribution matrix is ​​used to record the spatial distribution of multiple measurement positions and non-measurement positions in the numerical wave pool; multiple signal spectra are generated according to the multiple wave surface rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position; the multiple signal spectra and the spatial distribution matrix are used as a training data, and the pre-trained wave direction spectrum recognition model is trained using the training data to obtain a trained wave direction spectrum recognition model. The wave direction spectrum recognition model trained in this embodiment can be used for wave direction spectrum recognition, thereby effectively realizing intelligent recognition of wave direction spectrum recognition.

[0087] In the related art, numerical estimation methods of wave directional spectrum, such as maximum entropy method, maximum likelihood method and Bayesian method, are sensitive to the arrangement of wave measuring points, and the efficiency of the calculation process is low, resulting in unsatisfactory estimation accuracy and efficiency of directional spectrum in actual use. The technicians in this field fully consider the sparse characteristics of spatial position in wave field measurement and the data structure of wave directional spectrum, and are committed to developing a method using deep learning technology to improve the estimation accuracy and calculation efficiency of wave directional spectrum.

[0088] based on Figure 1 This embodiment proposes another method for generating a wave direction spectrum recognition model. In this method, when the pre-trained model is a pre-trained generative adversarial network, the pre-trained model includes a frequency wave direction estimation network and a discrimination network. At this time, step S104 may include:

[0089] The training data is input into the frequency wave direction estimation network to identify the wave direction spectrum and obtain the estimated wave direction spectrum;

[0090] According to the wave simulation test, the actual wave direction spectrum corresponding to the estimated wave direction spectrum is determined;

[0091] The estimated wave direction spectrum and the real wave direction spectrum are input into the discriminant network for calculation to obtain the credibility corresponding to the estimated wave direction spectrum;

[0092] The loss function value is determined according to the estimated wave direction spectrum, the real wave direction spectrum and the credibility, and the parameters in the pre-trained wave direction spectrum recognition model are updated according to the loss function value to obtain the trained wave direction spectrum recognition model.

[0093] It can be understood that, in this embodiment, relevant test data can be recorded in the wave simulation test, and the corresponding wave direction spectrum can be determined according to the test data, and can be used as the real wave direction spectrum.

[0094] It should be noted that this embodiment can be applied to the numerical wave pool scene to verify the wave directional spectrum corresponding to the test results when simulating three-dimensional short-peak waves, and can also be applied to the results of multiple single-point measurement devices at sea to identify and estimate the corresponding directional spectrum of the wave field in the sea area within the range.

[0095] Optionally, the frequency wave direction estimation network includes: a frequency domain identification module and a space identification module. At this time, the training data is input into the frequency wave direction estimation network to identify the wave direction spectrum to obtain the estimated wave direction spectrum, including:

[0096] Inputting multiple signal spectra in the training data into the frequency domain recognition module, so that the frequency domain recognition module performs energy distribution recognition according to the multiple signal spectra and the position identifiers marked on each signal spectrum, and obtains a target feature vector; wherein the target feature vector includes the distribution characteristics of wave energy at different frequencies;

[0097] The target feature vector and the spatial distribution matrix are input into the spatial recognition module for directional distribution recognition to obtain the target matrix; wherein the target matrix includes the distribution characteristics of wave energy in different directions within each frequency domain;

[0098] The target matrix is ​​determined to be the estimated wave direction spectrum.

[0099] Specifically, this embodiment can construct a generative adversarial network as a pre-trained wave direction spectrum recognition model, and the generative adversarial network can include two parts: a frequency wave direction estimation network and a discriminant network. Among them, the frequency wave direction estimation network can include two parts: a frequency domain recognition module and a space recognition module. In this embodiment, multiple signal spectra obtained by calculating multiple wave surface rise time series data can be used as input vectors, input into the input layer of the frequency domain recognition module, and the corresponding output vector, i.e., the target feature vector, is obtained after being processed by the frequency domain recognition module. Then, the target feature vector and the 0-1 spatial distribution matrix corresponding to the wave height meter are used as inputs, and input into the input layer of the space recognition module. After calculation and recognition by the space recognition module, a preliminary wave spectrum estimation result, i.e., an estimated wave direction spectrum, is obtained, and the real wave direction spectrum obtained when the numerical wave pool is calculated is input into the discriminant network, thereby realizing the authenticity identification of the wave direction discrimination result.

[0100] It should be noted that the target variable of the wave directional spectrum is a discrete combination of frequency and azimuth. Any frequency spectrum form can be expressed as a directional spectrum matrix expressed by a discrete combination of frequency and azimuth. The dimension of the matrix depends on the discretization of the wave directional spectrum in frequency and azimuth.

[0101] Optionally, the frequency domain recognition module includes a first convolutional layer, a first fully connected layer, and a first output layer. At this time, the multiple signal spectra in the training data are input into the frequency domain recognition module, so that the frequency domain recognition module performs energy distribution recognition according to the multiple signal spectra and the position identifiers marked on each signal spectrum, and obtains the target feature vector, including:

[0102] Inputting multiple signal spectra in the training data into the first convolutional layer, so that the first convolutional layer sequentially performs frequency domain local feature extraction and spectrum information complex feature extraction according to the multiple signal spectra and the position identifiers marked on each signal spectrum, to obtain spectrum information extraction features;

[0103] Inputting the spectral information extraction features into the first fully connected layer for nonlinear processing to obtain a corresponding first nonlinear processing result;

[0104] The nonlinear processing result is input into the first output layer for feature vector conversion to obtain the target feature vector.

[0105] Among them, the frequency domain recognition module may include a convolution layer, a fully connected layer and an output layer. During the training process, the present embodiment may first normalize the wave spectrum / cross spectrum obtained by the wave time calculation to obtain multiple normalized spectra and use them as signal spectra, and input the multiple signal spectra obtained into the convolution layer of the frequency domain recognition model, and then calculate the convolution result by sliding the convolution kernel in the convolution layer to extract the local features of the characteristic signal in the frequency domain, and then activate the function and the pooling layer to extract the complex features of the input spectrum information, and pass the result obtained after the convolution layer processing to the fully connected layer, and the fully connected layer processes the features extracted by the convolution layer into a feature vector, namely the target feature vector, through the output layer, which contains the main frequency features and energy distribution. The output result of the frequency domain recognition module represents the distribution of wave energy at different frequencies.

[0106] Optionally, the above-mentioned spatial recognition module includes a second convolutional layer, a second fully connected layer and a second output layer. At this time, the above-mentioned target feature vector and spatial distribution matrix are input into the spatial recognition module for direction distribution recognition to obtain the target matrix, including:

[0107] The target feature vector and the spatial distribution matrix are input into the second convolutional layer to extract spatial distribution features and phase distribution features in turn to obtain the corresponding spatial information extraction features;

[0108] Inputting the spatial information extraction features into the second fully connected layer for nonlinear processing to obtain a corresponding second nonlinear processing result;

[0109] The second nonlinear processing result is input into the second output layer for matrix conversion to obtain a target matrix; wherein each element in the target matrix is ​​the wave energy value of the corresponding frequency and wave direction.

[0110] Among them, the spatial recognition module may include a convolution layer, a fully connected layer and an output layer. During the training process, after the calculation of the frequency domain recognition module, the present embodiment can pass the output result of the frequency domain recognition module and the 0-1 spatial distribution matrix corresponding to the wave height meter as input to the convolution layer of the spatial recognition module. The convolution result is calculated under the sliding of the convolution kernel to extract the spatial distribution characteristics of the feature signal, followed by the operation of the activation function and the pooling layer to extract the complex characteristics of the input frequency domain distribution and spatial distribution information. After that, the present embodiment can pass the result of the convolution layer processing to the output layer of the spatial recognition module through the processing of the fully connected layer. The entire spatial recognition module combines the wave spectrum results obtained by the frequency domain recognition module and the processing of the phase information of the wave distributed in space, and obtains the distribution of wave energy in different directions within each frequency domain, thereby realizing the recognition process of the directional spectrum. After calculation and recognition by the spatial recognition module, its output is a preliminary wave spectrum estimation result, i.e., an estimated wave directional spectrum, which is in the form of a matrix, and each element of the matrix represents the wave energy value at a specific frequency and wave direction. Specifically, this output matrix has integrated the frequency domain information and spatial distribution characteristics, providing a relatively accurate preliminary estimate of the wave directional spectrum.

[0111] Specifically, this embodiment can combine the real wave direction spectrum (the data form is also a matrix) obtained by conducting experiments on the numerical wave pool, and input the real wave direction spectrum and the estimated wave direction spectrum into the discrimination network to realize the authenticity identification of the wave direction discrimination result. The discrimination network extracts features of the estimated wave direction spectrum and the real wave direction spectrum through multiple layers of convolution layers. These convolution layers extract the subtle features of the wave spectrum in frequency and wave direction respectively, and then compare the two sets of input features in the same feature space through the fully connected layer, and calculate the matrix error of the real wave direction spectrum and the estimated wave direction spectrum, that is, calculate the difference between the two matrices in each element to obtain the error matrix. Among them, the matrix of the real wave direction spectrum is derived from the calculation results in the numerical wave pool, representing the actual measured wave direction spectrum. Through multi-layer neural network calculation, the error matrix is ​​processed, and the discrimination probability, i.e., the reliability, is output.

[0112] Specifically, this embodiment can improve the estimation accuracy of the wave direction spectrum in the entire frequency and direction range by introducing a loss function related to credibility and combining it with the loss function of the generation network part.

[0113] Optionally, the above-mentioned determining the loss function value according to the estimated wave direction spectrum, the real wave direction spectrum and the credibility includes:

[0114] Inputting the estimated wave direction spectrum and the real wave direction spectrum into the first loss function for calculation, and obtaining the first loss function value output by the first loss function;

[0115] Inputting the credibility into the second loss function for calculation, and obtaining a second loss function value output by the second loss function;

[0116] The first loss function value and the second loss function value are added to obtain the loss function value.

[0117] Among them, the first loss function value is the loss function value corresponding to the process of frequency wave direction estimation network estimating wave direction spectrum, and the second loss function value is the loss function value corresponding to the process of discriminant network determining credibility.

[0118] Optionally, the first loss function value and the second loss function value are respectively:

[0119]

[0120] Among them, L MSE is the first loss function value, s and They represent the real wave direction spectrum and the estimated wave direction spectrum, L d is the second loss function value, Credibility.

[0121] Optionally, the target variable of the wave directional spectrum is a discrete combination of frequency and azimuth. Any spectrum form can be expressed as a directional spectrum matrix expressed by a discrete combination of frequency and azimuth. The dimension of the matrix depends on the discretization of the wave directional spectrum in frequency and azimuth. In this embodiment, the discrete parameter for the azimuth is N θ , the discrete number of frequencies is N ω , so frequency ω=0:2 / N ω :2rad / s, direction angle θ=-180:360 / N θ :180, that is, the total dimension of the wave direction spectrum target variable can be expressed as N ω ×N θ .

[0122] Optionally, this embodiment can carry out training on the constructed generative adversarial network, i.e., the pre-trained wave direction spectrum recognition model, and select the wave direction spectrum recognition model composed of the neural network parameters with the highest test accuracy according to the performance of the validation set and the test set during the training process.

[0123] like Figure 2 As shown, after generating the wave spectrum / cross spectrum results, i.e., multiple signal spectra, the present embodiment can input the multiple signal spectra into the frequency domain identification module 6 of the frequency wave direction estimation network 5 to obtain the target feature vector. Afterwards, the present embodiment can input the target feature vector and the wave height meter spatial distribution matrix 9 into the spatial identification module to obtain the wave spectrum estimation result 10 with directional angle information, i.e., the estimated wave direction spectrum, and then input the wave spectrum estimation result 10 and the corresponding real wave spectrum 11 into the discrimination network 12 to determine the authenticity / falseness of the wave direction discrimination result and obtain the credibility. Afterwards, the present embodiment can calculate the loss function according to the deviation between the estimated wave direction spectrum and the real wave direction spectrum, and optimize the hyperparameters of the entire frequency wave direction estimation network 5 according to the loss function. Among them, the present embodiment can perform supervised learning based on the calculated loss function and the optimization algorithm, and update the parameters of the neural network model until the loss function meets certain requirements or the number of training iterations is greater than the set value.

[0124] In this embodiment, the wave height meter wave surface rise time series of different spatial coordinates in the unknown short-peak wave field can be input into the frequency wave direction estimation network according to the trained neural network model. After the calculation of the neural network, the accurate and rapid recognition of the entire wave field direction spectrum is finally achieved. Figure 3 The wave direction spectrum in matrix form can be transformed into Figure 4 The general form of the wave direction spectrum is shown in Figure 1. Figure 3 and Figure 4 The wave direction spectrum has information such as frequency and direction angle, rad / s is the unit of frequency, and rad is the unit of direction angle.

[0125] This embodiment uses a generative adversarial network to deal with the spatial sparsity and wave frequency-wave direction identification problems in the wave field, thereby realizing the method of estimating and identifying the wave directional spectrum. There is no need to change the underlying theory of the wave directional spectrum, but to make improvements to the problem of unsatisfactory accuracy and efficiency of the wave directional spectrum caused by the low numerical calculation efficiency of the wave directional spectrum estimation. This embodiment combines a generative adversarial network and performs data-driven modeling based on the data characteristics of the wave directional spectrum, which effectively improves the accuracy of wave directional spectrum estimation and identification, as well as the forecast efficiency, and realizes intelligent identification of the wave directional spectrum.

[0126] In view of the large deviation of the estimation results of the traditional wave spectrum estimation algorithm in the arrangement of wave height meters in the wave field and the low calculation efficiency, this embodiment adopts a generative adversarial network recognition method, considers expressing the spatial position of a single monitoring point in the wave field in the form of a matrix, and based on the characteristics that the wave direction spectrum can also be expressed in the form of a frequency-direction angle matrix, a generative adversarial network based on a convolutional neural network is established.

[0127] The wave directional spectrum recognition model generation method proposed in this embodiment can perform wave directional spectrum recognition based on a generative adversarial network, and utilize the wave surface rise time series at multiple spatial measurement points in the wave field. Based on the frequency spectrum at different measurement points and the cross-spectrum of the wave surface rise at different spatial positions and other information, a data-driven model is established to quickly and accurately calculate and identify the directional spectrum of the frequency and wave direction information describing the wave components within the entire wave field, thereby improving the accuracy and efficiency of wave directional spectrum estimation.

[0128] like Figure 5 As shown, this embodiment proposes a wave direction spectrum recognition model generation device, which may include:

[0129] The determination unit 501 is used to determine a plurality of wave surface rise time series data corresponding to a plurality of measurement positions in the numerical wave pool, and to determine a spatial distribution matrix; wherein the wave surface rise time series data are obtained by measuring the wave surface rise at the corresponding measurement positions during the wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool;

[0130] A generating unit 502, configured to generate a plurality of signal spectra according to a plurality of wavefront rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position;

[0131] As unit 503, used for treating the plurality of signal spectra and spatial distribution matrices as a training data;

[0132] The training unit 504 is used to train the pre-trained wave direction spectrum recognition model using the training data to obtain a trained wave direction spectrum recognition model.

[0133] It should be noted that the processing of the determination unit 5015, the generation unit 502, the as unit 503 and the training unit 504 and the beneficial effects thereof can be respectively referred to in Figure 1 Steps S101 to S104 in the above are not described in detail.

[0134] Optionally, the determining unit 501 is further configured to:

[0135] The numerical wave pool is meshed to obtain multiple grids;

[0136] Generate a matrix to be assigned corresponding to the plurality of grids, wherein the matrix to be assigned includes elements to be assigned corresponding to the grids;

[0137] Determine a grid where each measurement position is located and use it as a first grid respectively; and determine a grid where each non-measurement position is located and use it as a second grid respectively;

[0138] In the matrix to be assigned, each element to be assigned corresponding to the first grid is assigned a value of 1, and each element to be assigned corresponding to the second grid is assigned a value of 0, so as to obtain a spatial distribution matrix.

[0139] Optionally, the generating unit 502 is further configured to:

[0140] For any wavefront rise time series data, a corresponding frequency spectrum and a first position identifier are generated according to the wavefront rise time series data, the first position identifier is used to identify the measurement position corresponding to the wavefront rise time series data, and the first position identifier is marked on the frequency spectrum;

[0141] For any two wavefront rise time series data, a corresponding cross spectrum and a second position identifier are generated according to the two wavefront rise time series data, the second position identifier is used to identify the measurement position corresponding to the two wavefront rise time series data, and the second position identifier is marked on the cross spectrum;

[0142] Normalizing all the frequency spectra and all the cross spectra to obtain a corresponding plurality of normalized spectra, each normalized spectrum being marked with a first position marker or a second position marker;

[0143] Each normalized spectrum is determined as a signal spectrum.

[0144] Optionally, when the pre-trained model is a pre-trained generative adversarial network, the pre-trained model includes a frequency wave direction estimation network and a discriminant network; the training unit 504 is further used to:

[0145] The training data is input into the frequency wave direction estimation network to identify the wave direction spectrum and obtain the estimated wave direction spectrum;

[0146] According to the wave simulation test, the actual wave direction spectrum corresponding to the estimated wave direction spectrum is determined;

[0147] The estimated wave direction spectrum and the real wave direction spectrum are input into the discriminant network for calculation to obtain the credibility corresponding to the estimated wave direction spectrum;

[0148] The loss function value is determined according to the estimated wave direction spectrum, the true wave direction spectrum and the credibility, and the parameters in the pre-trained wave direction spectrum recognition model are updated according to the loss function value.

[0149] Optionally, the frequency wave direction estimation network includes: a frequency domain identification module and a space identification module;

[0150] The training unit 504 is further configured to:

[0151] Inputting multiple signal spectra in the training data into the frequency domain recognition module, so that the frequency domain recognition module performs energy distribution recognition according to the multiple signal spectra and the position identifiers marked on each signal spectrum, and obtains a target feature vector; wherein the target feature vector includes the distribution characteristics of wave energy at different frequencies;

[0152] The target feature vector and the spatial distribution matrix are input into the spatial recognition module for directional distribution recognition to obtain the target matrix; wherein the target matrix includes the distribution characteristics of wave energy in different directions within each frequency domain;

[0153] The target matrix is ​​determined to be the estimated wave direction spectrum.

[0154] Optionally, the training unit 504 is further configured to:

[0155] Inputting the estimated wave direction spectrum and the real wave direction spectrum into the first loss function for calculation, and obtaining the first loss function value output by the first loss function;

[0156] Inputting the credibility into the second loss function for calculation, and obtaining a second loss function value output by the second loss function;

[0157] The first loss function value and the second loss function value are added to obtain the loss function value.

[0158] Optionally, the first loss function value and the second loss function value are respectively:

[0159]

[0160] Among them, L MSE is the first loss function value, s and They represent the real wave direction spectrum and the estimated wave direction spectrum, L d is the second loss function value, Credibility.

[0161] The wave direction spectrum recognition model generation device proposed in this embodiment can determine multiple wave surface rise time series data corresponding to multiple measurement positions in the numerical wave pool, and determine the spatial distribution matrix; wherein the wave surface rise time series data is obtained by measuring the wave surface rise at the corresponding measurement position during the wave simulation test in the numerical wave pool; the spatial distribution matrix is ​​used to record the spatial distribution of multiple measurement positions and non-measurement positions in the numerical wave pool; multiple signal spectra are generated according to the multiple wave surface rise time series data; wherein each signal spectrum corresponds to a measurement position, and each signal spectrum is marked with a position identifier for identifying the corresponding measurement position; the multiple signal spectra and the spatial distribution matrix are used as a training data, and the pre-trained wave direction spectrum recognition model is trained using the training data to obtain a trained wave direction spectrum recognition model. The wave direction spectrum recognition model trained in this embodiment can be used for wave direction spectrum recognition, thereby effectively realizing intelligent recognition of wave direction spectrum recognition.

[0162] The wave direction spectrum recognition model generation 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.

[0163] The embodiment of the present invention also provides a computer device having the above Figure 5 The wave direction spectrum recognition model generating device shown.

[0164] See also Figure 6 , 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 6 A processor 10 is taken as an example.

[0165] 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.

[0166] 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.

[0167] 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.

[0168] 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.

[0169] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0170] 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.

[0171] 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 method for generating a wave direction spectrum recognition model, characterized in that: include: Determine a plurality of wave surface rise time series data corresponding to a plurality of measurement positions in the numerical wave pool, and determine a spatial distribution matrix; wherein the wave surface rise time series data is obtained by measuring the wave surface rise at the corresponding measurement positions during a wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool; Generate multiple signal spectra according to the multiple wavefront rise time series data; wherein each of the signal spectra corresponds to the measurement position, and each of the signal spectra is marked with a position identifier for identifying the corresponding measurement position; The plurality of signal spectra and the spatial distribution matrix are used as a training data as a whole, and the pre-trained wave direction spectrum recognition model is trained using the training data to obtain a trained wave direction spectrum recognition model; Wherein, determining the spatial distribution matrix includes: Gridding the numerical wave pool to obtain a plurality of grids; Generate a matrix to be assigned corresponding to the plurality of grids, wherein the matrix to be assigned includes elements to be assigned corresponding to the grids; Determine the grid where each of the measurement positions is located and use them as a first grid respectively; and determine the grid where each of the non-measurement positions is located and use them as a second grid respectively; In the matrix to be assigned, assigning a value of 1 to each element to be assigned corresponding to the first grid, and assigning a value of 0 to each element to be assigned corresponding to the second grid, so as to obtain the spatial distribution matrix; The step of generating a plurality of signal spectra according to the plurality of wavefront rise time series data comprises: For any of the wavefront rise time series data, generate a corresponding frequency spectrum and a first position identifier according to the wavefront rise time series data, wherein the first position identifier is used to identify the measurement position corresponding to the wavefront rise time series data, and mark the first position identifier on the frequency spectrum; For any two of the wavefront rise time series data, generate a corresponding cross spectrum and a second position identifier according to the two wavefront rise time series data, the second position identifier is used to identify the measurement position corresponding to the two wavefront rise time series data, and mark the second position identifier on the cross spectrum; Normalizing all the frequency spectra and all the cross spectra to obtain a corresponding plurality of normalized spectra, each of the normalized spectra being marked with the first position marker or the second position marker; Determining each of the normalized spectra as a signal spectrum; Wherein, when the pre-trained model is a pre-trained generative adversarial network, the pre-trained model includes a frequency wave direction estimation network and a discrimination network; The using the training data to train the pre-trained wave direction spectrum recognition model comprises: Inputting the training data into the frequency wave direction estimation network to identify the wave direction spectrum and obtain an estimated wave direction spectrum; Determining a real wave direction spectrum corresponding to the estimated wave direction spectrum according to the wave simulation test; Inputting the estimated wave direction spectrum and the real wave direction spectrum into the discrimination network for calculation to obtain the credibility corresponding to the estimated wave direction spectrum; Determining a loss function value according to the estimated wave direction spectrum, the true wave direction spectrum and the credibility, and updating parameters in the pre-trained wave direction spectrum recognition model according to the loss function value; Wherein, the frequency wave direction estimation network includes: a frequency domain identification module and a space identification module; The step of inputting the training data into the frequency wave direction estimation network to identify the wave direction spectrum and obtain the estimated wave direction spectrum comprises: Inputting the multiple signal spectra in the training data into the frequency domain identification module, so that the frequency domain identification module performs energy distribution identification according to the multiple signal spectra and the position identifier marked on each of the signal spectra to obtain a target feature vector; wherein the target feature vector includes distribution characteristics of wave energy at different frequencies; Input the target feature vector and the spatial distribution matrix into the spatial identification module for directional distribution identification to obtain a target matrix; wherein the target matrix includes the distribution characteristics of wave energy in different directions within each frequency domain; The target matrix is ​​determined as the estimated wave direction spectrum.

2. The method according to claim 1, characterized in that The determining of the loss function value according to the estimated wave direction spectrum, the real wave direction spectrum and the credibility comprises: Inputting the estimated wave direction spectrum and the real wave direction spectrum into a first loss function for calculation, to obtain a first loss function value output by the first loss function; Inputting the credibility into a second loss function for calculation to obtain a second loss function value output by the second loss function; The first loss function value and the second loss function value are added to obtain the loss function value.

3. The method according to claim 2, characterized in that The first loss function value and the second loss function value are respectively: ； ； in, L MSE is the first loss function value, and represent the true wave direction spectrum and the estimated wave direction spectrum respectively, L d is the second loss function value, For the credibility.

4. A wave direction spectrum recognition model generation device, characterized in that: include: A determination unit, used to determine a plurality of wave surface rise time series data corresponding to a plurality of measurement positions in a numerical wave pool, and to determine a spatial distribution matrix; wherein the wave surface rise time series data are obtained by measuring the wave surface rise at the corresponding measurement positions during a wave simulation test in the numerical wave pool; and the spatial distribution matrix is ​​used to record the spatial distribution of the plurality of measurement positions and non-measurement positions in the numerical wave pool; A generating unit, configured to generate a plurality of signal spectra according to the plurality of wavefront rise time series data; wherein each of the signal spectra corresponds to the measurement position, and each of the signal spectra is marked with a position identifier for identifying the corresponding measurement position; As a unit, used to treat the plurality of signal spectra and the spatial distribution matrix as a training data as a whole; A training unit, used for training the pre-trained wave direction spectrum recognition model using the training data to obtain a trained wave direction spectrum recognition model; Wherein, the determining unit is further used for: Gridding the numerical wave pool to obtain a plurality of grids; Generate a matrix to be assigned corresponding to the plurality of grids, wherein the matrix to be assigned includes elements to be assigned corresponding to the grids; Determine the grid where each of the measurement positions is located and use them as a first grid respectively; and determine the grid where each of the non-measurement positions is located and use them as a second grid respectively; In the matrix to be assigned, assigning a value of 1 to each element to be assigned corresponding to the first grid, and assigning a value of 0 to each element to be assigned corresponding to the second grid, so as to obtain the spatial distribution matrix; Wherein, the generating unit is further used for: For any of the wavefront rise time series data, generate a corresponding frequency spectrum and a first position identifier according to the wavefront rise time series data, wherein the first position identifier is used to identify the measurement position corresponding to the wavefront rise time series data, and mark the first position identifier on the frequency spectrum; For any two of the wavefront rise time series data, generate a corresponding cross spectrum and a second position identifier according to the two wavefront rise time series data, the second position identifier is used to identify the measurement position corresponding to the two wavefront rise time series data, and mark the second position identifier on the cross spectrum; Normalizing all the frequency spectra and all the cross spectra to obtain a corresponding plurality of normalized spectra, each of the normalized spectra being marked with the first position marker or the second position marker; Determining each of the normalized spectra as a signal spectrum; Wherein, when the pre-trained model is a pre-trained generative adversarial network, the pre-trained model includes a frequency wave direction estimation network and a discrimination network; the training unit is further used for: Inputting the training data into the frequency wave direction estimation network to identify the wave direction spectrum and obtain an estimated wave direction spectrum; Determining a real wave direction spectrum corresponding to the estimated wave direction spectrum according to the wave simulation test; Inputting the estimated wave direction spectrum and the real wave direction spectrum into the discrimination network for calculation to obtain the credibility corresponding to the estimated wave direction spectrum; Determining a loss function value according to the estimated wave direction spectrum, the true wave direction spectrum and the credibility, and updating parameters in the pre-trained wave direction spectrum recognition model according to the loss function value; The frequency wave direction estimation network includes: a frequency domain identification module and a space identification module; the training unit is also used for: Inputting the multiple signal spectra in the training data into the frequency domain identification module, so that the frequency domain identification module performs energy distribution identification according to the multiple signal spectra and the position identifier marked on each of the signal spectra to obtain a target feature vector; wherein the target feature vector includes distribution characteristics of wave energy at different frequencies; Input the target feature vector and the spatial distribution matrix into the spatial identification module for directional distribution identification to obtain a target matrix; wherein the target matrix includes the distribution characteristics of wave energy in different directions within each frequency domain; The target matrix is ​​determined as the estimated wave direction spectrum.

5. 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 direction spectrum recognition model generation method according to any one of claims 1 to 3 by executing the computer instructions.

6. 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 direction spectrum recognition model generation method according to any one of claims 1 to 3.

Images